The reviewed record of science sign in
Pith

arxiv: 2607.06224 · v1 · pith:UVKNUDPZ · submitted 2026-07-07 · cs.LG

Canopy: A Heterograph Foundation Model for Metabolic Engineering

Jake Bowden , Laurence Legon , Satnam Surae This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-07-08 12:42 UTCglm-5.2pith:UVKNUDPZrecord.jsonopen to challenge →

classification cs.LG
keywords heterogeneous graph neural networkmetabolic engineeringfermentation titer predictionknowledge graphself-supervised pretraininggraph foundation modelmulti-modal representation learningstrain design
0
0 comments X

The pith

Graph foundation model nearly doubles titer prediction over tabular ML

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper argues that a pretrained heterogeneous graph neural network, operating over a unified knowledge graph spanning genes, proteins, metabolites, reactions, pathways, strains, and fermentation experiments across organisms, can learn representations that substantially improve prediction of fermentation product titers. The authors build a 6.9-million-node knowledge graph integrating ten data sources, encode each node type with domain-specific foundation models (ESM-2 for proteins, MoLFormer for chemicals, PubMedBERT for text), and pretrain a Heterogeneous Graph Transformer with four self-supervised objectives. On held-out fermentation experiments, frozen graph embeddings fed to a lightweight probe achieve R²=0.41, compared to R²=0.24 for the best tabular baseline and lower scores for homogeneous graph variants. The central claim is that the relational structure of a cross-organism, multi-modal knowledge graph encodes information useful for predicting biological production outcomes that flat feature vectors and single-organism graph models cannot capture.

Core claim

The paper's central result is that a heterogeneous graph transformer pretrained on a multi-modal metabolic-engineering knowledge graph produces frozen embeddings that predict fermentation titer with R²=0.41 using only a lightweight probe, outperforming tabular baselines (best R²=0.24), homogeneous GraphSAGE (R²=0.24), and vanilla HGT without architectural augmentations (R²=0.31). The improvement holds across model scales from 80M to 3B parameters, with the largest model achieving the best R² and AUROC. Ablations show that virtual nodes contribute the most among architectural augmentations (removing them drops R² by 10.3 points), that learned uncertainty weighting of the four pretraining loss

What carries the argument

The central object is Canopy's heterogeneous knowledge graph: 6.9 million nodes across 13 types (metabolites, reactions, UniRef protein clusters, InterPro domains, genes, chassis organisms, strains, pathways, GO terms, taxa, experiments, transcriptomic measurements) connected by 34 typed edge relations. Node features are dispatched by property prefix to frozen foundation-model encoders — ESM-2 (650M) for protein sequences, MoLFormer-XL for SMILES strings, PubMedBERT for biomedical text — and combined with SignNet Laplacian positional encodings and random-walk structural encodings. The encoder is a stack of Heterogeneous Graph Transformer convolution layers with per-type and per-relation注意力,

If this is right

  • If the graph structure genuinely drives the improvement, the same pretrained embeddings could serve as conditioning for generative strain-design pipelines — the authors describe pairing them with flow-matching models and Bayesian optimization loops to propose multi-gene interventions scored by the frozen titer probe.
  • The 4,791-experiment benchmark with deterministic splits addresses a gap in metabolic-engineering ML evaluation: no existing resource combines experimental titer measurements, cross-organism coverage, and multi-omic context with held-out evaluation, which could make it a standard test bed if released.
  • Cross-organism transfer becomes testable: the heterogeneous graph naturally links homologs, pathways, and compounds across species, so pretraining on well-studied organisms and evaluating on underexplored ones is a direct extension the authors propose.
  • The modest scaling gain from 500M to 3B parameters (R² 0.38→0.41) suggests the current bottleneck is experimental data volume, not model capacity, implying that data expansion — not larger models — is the limiting factor for further improvement.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The missing graph-free control (concatenating frozen ESM-2 and MoLFormer embeddings without any graph structure) is the critical falsifier: if an MLP on pooled encoder features matches or approaches R²=0.41, the heterogeneous graph transformer is not the source of the improvement, and the result reduces to 'foundation-model features are good for titer prediction.'
  • The shallow model (L=2) outperforming the deep model (L=6) in ablation suggests oversmoothing is already active at modest depth on this graph, which raises the question of whether the 3B model's advantage comes from depth-related representation quality or simply from wider per-head dimensions.
  • If the AUROC task is partly trivialised by fermentation-volume correlation (AUROC 0.65 from volume alone), the regression R² is the more trustworthy metric, and the binary classification gains should be interpreted cautiously.
  • The virtual-node ablation's large effect (−10.3 R²) may indicate that graph diameter reduction matters more than heterogeneous message passing on this particular graph, which would reframe the contribution from 'heterogeneous modeling' to 'graph connectivity engineering.'

Load-bearing premise

The claim that heterogeneous graph structure drives the titer-prediction improvement rests on comparisons against tabular baselines and homogeneous graph variants, but the paper has not yet run the decisive control: feeding the same frozen ESM-2 and MoLFormer embeddings into an MLP without any graph structure. Without that ablation, the improvement could come from the pretrained feature encoders rather than from the graph transformer.

What would settle it

Concatenate frozen ESM-2 and MoLFormer embeddings of each strain's genes and target compound into a flat vector, apply the same MLP probe, and use the same train/test split. If this graph-free probe matches or exceeds R²=0.41, the heterogeneous graph transformer is not the source of Canopy's improvement.

Figures

Figures reproduced from arXiv: 2607.06224 by Jake Bowden, Laurence Legon, Satnam Surae.

Figure 1
Figure 1. Figure 1: CANOPY heterograph schema. Each node type has a distinct shape and is coloured by data domain (genomics, proteomics, metabolomics, experiment, ontology); each edge is typed by relation. Strain-to-gene modification edges include knockouts, knockins, and knockdowns. 3. Method 3.1. Knowledge Graph Construction CANOPY’s knowledge graph is constructed using Bio￾Cypher (Lobentanzer et al., 2023). We implement te… view at source ↗
Figure 2
Figure 2. Figure 2: Input feature construction. Each node type τ (v) has its own node emb, sign net (consuming a 10-d Laplacian eigenvec￾tor positional encoding), and rw emb (consuming a 16-d random￾walk positional encoding); the three streams are summed to form the per-node residual stream h (0) v . mean-pooled across residues. Chemical structures (smi ) are encoded by MoLFormer-XL (Ross et al., 2021). Biomed￾ical text (text… view at source ↗
Figure 3
Figure 3. Figure 3: One HGT layer. The dst-side query Qτ(t) is type-specific; keys Kτ(s) and values Vτ(s) are projected per source type, while attention logits and messages are scaled by per-relation matrices WATT ϕ , WMSG ϕ and a learnable relation prior µ⟨τ(s),ϕ,τ(t)⟩. Attention is softmaxed over the destination’s neighbourhood and aggregated; a per-type output projection A-Linτ(t) feeds an FFN sub-block. Both attention and… view at source ↗
Figure 4
Figure 4. Figure 4: Readout. After L stacked HGT layers, JumpingKnowl￾edge max-pools the per-layer representations {h (ℓ) v } (combined with a learnable input-residual scale α (in) [τ] ) and a final per-type Lay￾erNorm produces the encoder output h out v . (src, dst) pairs from the corresponding node pools within the sampled subgraph and rejecting pairs that are already edges. Masked node modelling. We mask 15% of node featur… view at source ↗
Figure 5
Figure 5. Figure 5: Pretraining objective. Four parallel heads consume the encoder output {zv}v∈V : link prediction (BCE), per-type masked node modelling (MSE), graph-distance regression (MSE), and an Experiment↔Experiment contrastive loss (MSE on cosine similarity). Their weighted sum, using Kendall et al. (2018) uncertainty weighting, forms the training objective. BFS expansion to prevent hub-dominated subgraphs but are inc… view at source ↗
read the original abstract

Designing microbial strains that produce high-value chemicals at commercially viable titers remains a central challenge in metabolic engineering. Existing computational approaches either rely on stoichiometric constraint-based models that cannot learn from experimental data, or apply tabular machine learning to hand-crafted features that discard the relational structure of biological knowledge. We present Canopy, a heterogeneous graph foundation model that integrates ten public and proprietary data sources into a unified knowledge graph (KG) of 6.9M nodes across 13 types and 34 edge types, covering genes, proteins, metabolites, reactions, pathways, strains, and fermentation experiments. Node features are encoded through domain-specific foundation models (ESM-2 for protein sequences, MoLFormer for chemical SMILES, and PubMedBERT for biomedical text), yielding a multi-modal representation within a single graph. We pretrain a Heterogeneous Graph Transformer (HGT) augmented with SignNet positional encodings, Jumping Knowledge aggregation, and virtual nodes using four self-supervised objectives (link prediction, masked node modelling, distance prediction, and contrastive experiment clustering), balanced via learned homoscedastic uncertainty weighting. On the downstream task of fermentation titer prediction, frozen Canopy embeddings achieve $R^{2} = 0.41$ with a lightweight probe, outperforming tabular baselines (best $R^{2} = 0.24$) and homogeneous GNN variants.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 9 minor

Summary. The paper presents Canopy, a heterogeneous graph foundation model for metabolic engineering. The authors construct a knowledge graph of 6.9M nodes (13 types) and 34 edge types from ten data sources, encode node features using frozen pretrained models (ESM-2, MoLFormer, PubMedBERT), and pretrain a Heterogeneous Graph Transformer (HGT) with four self-supervised objectives (link prediction, masked node modeling, distance prediction, contrastive experiment clustering) combined via learned uncertainty weighting. On fermentation titer prediction using 4,791 experiments with a deterministic 5-fold CV split, frozen Canopy embeddings achieve R²=0.41 with a lightweight probe, outperforming tabular baselines (best R²=0.24) and homogeneous GNN variants. Ablations test pretraining objectives, architectural components, depth, and loss weighting.

Significance. The integration of a multi-modal metabolic-engineering knowledge graph at this scale is a genuine contribution to an underserved domain. The system design—combining BioCypher-based KG construction, schema-driven multi-modal feature dispatch, HGT with SignNet positional encodings and Jumping Knowledge, and multi-task self-supervised pretraining—is well-executed and technically sound. The deterministic MD5-hashed split protocol and the hold-out integrity design (excluding held-out Experiment nodes from all pretraining supervision while retaining them in the message-passing graph) are commendable. The ablation suite (Tables 4–7) is thorough in testing individual components. The authors are transparent about limitations, including the missing graph-free encoder control and data sparsity.

major comments (3)
  1. §4.2, Table 3: The central claim that graph structure drives the R²=0.41 improvement rests on the comparison between Canopy and tabular baselines (R²=0.24). However, the tabular baselines operate on a 429-dim raw experiment-condition vector, while Canopy's Experiment node aggregates ESM-2, MoLFormer, and PubMedBERT features from neighboring graph nodes. This conflates two factors: (a) the heterogeneous graph transformer and (b) access to pretrained foundation-model features. The authors acknowledge this in §5 (Limitations, third point) as a planned ablation. Without a graph-free control that concatenates frozen ESM-2 and MoLFormer embeddings of a strain's genes and target compound and applies an MLP probe, the attribution of improvement to graph structure rather than to the feature encoders cannot be established. This is load-bearing for the paper's core claim. The graph-internal basings
  2. §4.2, Table 3: No error bars or standard deviations are reported across the 5 CV folds (n=410 test experiments). The R²=0.41 vs. 0.24 comparison lacks any statistical significance assessment. Given the modest absolute R² values and the relatively small test set, reporting per-fold variance and a significance test would strengthen the claim that the improvement is robust rather than an artifact of split variance.
  3. §5, Limitations, fourth point: The 4,791-experiment benchmark and split files are stated to be released 'in a forthcoming publication,' and trained model weights and LIMS records are not released. For a paper claiming a foundation model and a benchmark, the inability to reproduce any of the headline results—no code, no data, no weights—is a significant gap. At minimum, the benchmark split files and probe evaluation code should be released to allow independent verification of the R²=0.41 claim.
minor comments (9)
  1. Table 4: The '∆pp' column header is ambiguous—it appears to mean percentage-point change, but this should be stated explicitly. The row showing '✓ ✓ ✓ ✓' with R²=0.359 and no ∆pp value is the reference; labeling it as such would improve clarity.
  2. Table 7: The flat-weighting row reports '∆R² = −9.18', which appears to be in percentage points, but the learned-weighting row reports R²=0.359 with no ∆. Consistent units (decimal vs. percentage points) across tables would aid comparison.
  3. §3.4, contrastive experiment-pair loss: The target similarity s_ij = 1 − d_ij/d_max is defined, but d_max is not specified. Is it the maximum graph distance within the batch, the subgraph, or the full graph? This should be clarified.
  4. §3.5: The multi-anchor sampling strategy rebalances GenomicGene fraction 'from 78% to 15%'. It would help to state what the Experiment-node coverage actually is (the '13×' relative figure is given but the absolute percentage is not).
  5. §4.3, Table 6: The depth ablation compares L=2 (h=224) vs. L=6 (h=128) at 500M scale. The authors note the shallow model performs better (R²=0.370 vs. 0.359) but retain the deeper default. The justification ('the gap is small and the deeper topology matches the 3B configuration') is reasonable but could note whether the 3B model itself was tested at L=2.
  6. Figure references: Figures 2–5 are referenced in §3.3 but the figure captions themselves do not include figure numbers in the text provided, making cross-referencing difficult. Ensure figures are numbered consistently.
  7. §3.3: The virtual node type is described as 'added with bidirectional edges to all other nodes.' Given the subgraph sampling approach (§3.5), it would be useful to clarify whether virtual node edges are added within each sampled subgraph or at the full-graph level before sampling.
  8. Table 1: The total node count is listed as 6,863,526, but the abstract states 6.9M. Minor rounding inconsistency—consider aligning.
  9. §4.1: The hyperparameter sweep used ~1,300 XPU-hours total (500 + 800). It would be useful to state the total compute including the final training runs at each scale.

Simulated Author's Rebuttal

3 responses · 2 unresolved

We thank the referee for a careful and constructive report. The referee correctly identifies that our central claim—graph structure drives the R²=0.41 improvement—requires a graph-free foundation-model control to disentangle the contribution of the heterogeneous graph transformer from the contribution of the frozen ESM-2/MoLFormer/PubMedBERT feature encoders. We agree this is load-bearing and will add the control. We also agree on error bars and will release benchmark split files and probe evaluation code. One genuine constraint remains: trained model weights and proprietary LIMS records cannot be released.

read point-by-point responses

  1. Referee: §4.2, Table 3: The central claim that graph structure drives the R²=0.41 improvement rests on the comparison between Canopy and tabular baselines (R²=0.24). However, the tabular baselines operate on a 429-dim raw experiment-condition vector, while Canopy's Experiment node aggregates ESM-2, MoLFormer, and PubMedBERT features from neighboring graph nodes. This conflates two factors: (a) the heterogeneous graph transformer and (b) access to pretrained foundation-model features. The authors acknowledge this in §5 (Limitations, third point) as a planned ablation. Without a graph-free control that concatenates frozen ESM-2 and MoLFormer embeddings of a strain's genes and target compound and applies an MLP probe, the attribution of improvement to graph structure rather than to the feature encoders cannot be established. This is load-bearing for the paper's core claim.

    Authors: The referee is correct that the current Table 3 comparison conflates two factors: the heterogeneous graph transformer and access to pretrained foundation-model features. We agree this is load-bearing for the core claim and will add the requested graph-free control in the revision. Specifically, we will implement an MLP probe on the concatenation of frozen ESM-2 mean-pooled embeddings (for each strain's gene set), frozen MoLFormer embeddings (for the target compound SMILES), and frozen PubMedBERT embeddings (for experiment text), using the same MD5-hashed 5-fold CV split. This control isolates whether the graph structure and message passing contribute beyond what the foundation-model encoders alone provide. We note that the existing graph-internal ablations in Table 3 (GraphSAGE vanilla at R²=0.241, HGT vanilla at R²=0.308, HGT+SN/JK/VN at R²=0.413) already provide partial evidence that graph architecture matters—these baselines share the same frozen encoder features but differ in backbone, and the improvement from 0.241 to 0.413 tracks architectural choices. However, we concede that none of these baselines controls for the foundation-model encoders in the way the referee describes, and the graph-free encoder control is the cleaner and more direct test. We will add it as a new row in Table 3 and revise the claims in §4.2 and the Abstract accordingly. revision: yes

  2. Referee: §4.2, Table 3: No error bars or standard deviations are reported across the 5 CV folds (n=410 test experiments). The R²=0.41 vs. 0.24 comparison lacks any statistical significance assessment. Given the modest absolute R² values and the relatively small test set, reporting per-fold variance and a significance test would strengthen the claim that the improvement is robust rather than an artifact of split variance.

    Authors: We agree. We will report per-fold R² values, standard deviations across the 5 folds, and a paired statistical test (Wilcoxon signed-rank on per-fold R² values, given n=5 folds) for all methods in Table 3 in the revised manuscript. We acknowledge that with only 5 folds the statistical power of any significance test is limited, but reporting the variance is important for the reader to assess robustness. We will also add a note discussing the limitations of inference at this sample size. revision: yes

  3. Referee: §5, Limitations, fourth point: The 4,791-experiment benchmark and split files are stated to be released 'in a forthcoming publication,' and trained model weights and LIMS records are not released. For a paper claiming a foundation model and a benchmark, the inability to reproduce any of the headline results—no code, no data, no weights—is a significant gap. At minimum, the benchmark split files and probe evaluation code should be released to allow independent verification of the R²=0.41 claim.

    Authors: We agree that the release scope described in the current manuscript is insufficient. In the revision we will release (1) the 4,791-experiment benchmark metadata with the deterministic MD5-hashed 5-fold CV split files, and (2) the probe evaluation code (linear and MLP probe training, R²/RMSE/Spearman computation) as a public repository. This will allow independent verification of the R²=0.41 claim given the released split and probe code. We will also release the subgraph sampling configuration and the schema YAML so that the knowledge graph construction pipeline is reproducible from public data sources. The trained model weights and proprietary LIMS records will remain constrained—the LIMS data is proprietary to our company and cannot be released, and the model weights encode learned representations over that proprietary data. We will state this constraint explicitly and adjust the 'foundation model' framing to be transparent about what is and is not reproducible from public components alone. revision: partial

standing simulated objections not resolved
  • Trained model weights cannot be released because they encode learned representations over proprietary LIMS fermentation records. This is a genuine constraint of the industrial setting and cannot be resolved for this submission.
  • Proprietary LIMS DBTL records (a subset of the 4,791 experiments) cannot be released. The literature-mined subset and split files will be released, but the in-house experimental records will remain constrained.

Circularity Check

0 steps flagged

No significant circularity: pretraining objectives are self-supervised on graph structure, downstream probe is trained on held-out experiments excluded from all pretraining supervision, and the deterministic split is applied consistently.

full rationale

The paper's central claim is that frozen Canopy embeddings achieve R²=0.41 on titer prediction, outperforming tabular baselines (R²=0.24). Walking the derivation chain: (1) The four pretraining objectives (link prediction, masked node modeling, distance prediction, contrastive experiment clustering) are self-supervised on graph structure and do not fit titer values. (2) The downstream probe is trained on held-out Experiment nodes explicitly excluded from all pretraining supervision: 'edges incident to any val/test Experiment are excluded from the link-prediction label set, masked-node-modelling ignores held-out Experiment rows when sampling its mask, and the contrastive Experiment-clustering loss filters held-out Experiments out of its pair pool before sampling.' (3) The MD5-hashed split is applied consistently across pretraining and probing. (4) The tabular baselines use 'the same MD5-hashed train/test split as the CANOPY probe' and operate on 'the same input that feeds the Experiment node in the graph.' (5) The graph-internal comparisons (Canopy vs vanilla HGT vs GraphSAGE, Table 3) share the same node features, partially isolating the contribution of the heterogeneous graph transformer. The missing graph-free encoder control (acknowledged in Limitations) is a correctness/ablation gap, not a circularity issue: the paper does not define its inputs in terms of its outputs, nor does it fit a parameter to the target and call it a prediction. The self-citations present (Hu et al. 2020b for HGT, Kendall et al. 2018 for uncertainty weighting, Lim et al. 2022 for SignNet) are to external, independently published work and are not load-bearing for the central empirical claim. The only minor concern is that the Optuna hyperparameter sweep optimizes 'the held-out titer probe R² at its best epoch,' which means the reported R² numbers are selected from the best of 1,000+200 trials on the same validation signal — this is a mild overfitting risk but not circularity, since the test split is held out from both pretraining and probe fitting. No step in the derivation chain reduces to its inputs by construction.

Axiom & Free-Parameter Ledger

18 free parameters · 6 axioms · 1 invented entities

The model has approximately 18 free hyperparameters selected via Optuna sweeps optimizing held-out probe R², plus several hand-set constants (residual scale, contrastive temperature, subgraph budget, PE dimensions). The axioms are primarily domain assumptions about feature encoder quality, schema fidelity, sampling adequacy, split integrity, and titer data quality. No new biological entities are invented; the virtual node is a standard architectural technique.

free parameters (18)
  • hidden dimension (d_h) = 64/128/256 (Demo/500M/3B)
    Selected via Optuna sweep optimizing held-out probe R².
  • number of layers (L) = 6 (500M), 12 (3B)
    Selected via Optuna sweep; depth ablation shows L=2 outperforms L=6 but deeper config retained.
  • number of attention heads (H) = 4 (Demo/500M), 8 (3B)
    Selected via Optuna sweep.
  • FFN expansion ratio (r) = 4
    Selected from {2,4} via Optuna sweep.
  • learning rate = 2e-3 (Tier 1 winner)
    Log-uniform search [1e-4, 5e-3] via Optuna.
  • batch size = 256 (Tier 1 winner)
    Categorical search {8-256} via Optuna.
  • warmup epochs = 4 (Tier 1 winner)
    Integer search [1,10] via Optuna.
  • cosine end offset = 6 (Tier 1 winner)
    Integer search [1,30] via Optuna.
  • dropout = searched in [0.05, 0.40]
    Selected via Optuna sweep.
  • probe hidden dim = searched in {32,64,128,256}
    Selected via Optuna sweep optimizing probe R².
  • probe num layers = searched in [1,4]
    Selected via Optuna sweep.
  • learned log-variance per task (log σ_i) = converged values not reported
    Learnable uncertainty weighting parameters for the four pretraining losses (Eq. 2).
  • residual scale initial value = 0.1
    Hand-set initial value for scaled residual connections.
  • contrastive temperature (τ) = 0.07
    Hand-set for contrastive experiment-pair loss.
  • subgraph node budget (N_max) = 1000
    Hand-set default for multi-anchor sampling.
  • number of subgraphs sampled = 10000 (headline), 5000 (ablations)
    Hand-set training budget.
  • Laplacian eigenvector count (k) = 10
    Dimensionality of SignNet positional encoding.
  • RWPE dimension = 16
    Dimensionality of random-walk positional encoding.
axioms (6)
  • domain assumption ESM-2, MoLFormer-XL, and PubMedBERT produce meaningful frozen feature representations for protein sequences, chemical structures, and biomedical text respectively.
    Section 3.2: these pretrained models are used as frozen feature extractors; their quality directly determines node feature quality.
  • domain assumption The heterogeneous graph schema (13 node types, 34 edge types) faithfully represents the biological relationships relevant to fermentation titer prediction.
    Section 3.1: the schema design determines what information the model can access.
  • domain assumption Subgraph sampling via multi-anchor BFS preserves sufficient graph context for representation learning.
    Section 3.5: training on subgraphs of N_max=1000 nodes approximates full-graph message passing.
  • domain assumption The MD5-hashed experiment-node split produces train/test sets without distributional shift.
    Section 4.1: the split is deterministic but not stratified by compound, organism, or titer distribution; the authors note a per-fold check is planned.
  • standard math Homoscedastic uncertainty weighting (Kendall et al. 2018) appropriately balances the four pretraining losses.
    Section 3.4, Eq. 2: the loss combination scheme is a standard multi-task learning technique.
  • domain assumption Literature-mined fermentation titer values are accurate and comparable across studies.
    Section 3.6: titers from published studies and LIMS are treated as ground truth; cross-study variability in fermentation conditions is not explicitly controlled.
invented entities (1)
  • Virtual node type no independent evidence
    purpose: Bidirectional edges to all other nodes to reduce graph diameter and provide shortcut paths between disconnected components.
    Section 3.3: a standard GNN technique (Gilmer et al. 2017), not a new biological entity. No falsifiable biological prediction; ablation in Table 5 shows removing it drops R² by 10.3pp.

pith-pipeline@v1.1.0-glm · 22071 in / 4096 out tokens · 568281 ms · 2026-07-08T12:42:56.124973+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

65 extracted references · 65 canonical work pages · 10 internal anchors

  1. [1]

    Heterogeneous Graph Transformer

    Ziniu Hu and Yuxiao Dong and Kuansan Wang and Yizhou Sun , title =. CoRR , volume =. 2020 , url =. 2003.01332 , timestamp =

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  2. [2]

    Bioinformatics Advances , author =

    Knowledge graph embeddings in the biomedical domain: are they useful?. Bioinformatics Advances , author =. 2024 , pages =. doi:10.1093/bioadv/vbae097 , abstract =

    work page doi:10.1093/bioadv/vbae097 2024

  3. [3]

    Enhancing link prediction in biomedical knowledge graphs with

    Hu, Emy Yue and Oleshko, Svitlana and Firmani, Samuele and Cheng, Hui and Zhu, Zhaocheng and Ulmer, Maria and Arnold, Matthias and Colomé-Tatché, Maria and Tang, Jian and Xhonneux, Sophie and Marsico, Annalisa , month = jan, year =. Enhancing link prediction in biomedical knowledge graphs with. Nature Biomedical Engineering , publisher =. doi:10.1038/s415...

    work page doi:10.1038/s41551-025-01598-z

  4. [4]

    doi:10.64898/2026.02.20.707076 , abstract =

    Zhang, Xinmeng and Jeong, Eugene and Yan, Chao and Feng, Yubo and Lyu, Linshuoshuo and Guo, Xingyi and Chen, You , month = feb, year =. doi:10.64898/2026.02.20.707076 , abstract =

    work page doi:10.64898/2026.02.20.707076 2026

  5. [5]

    National Science Review , author =

    Cell-. National Science Review , author =. 2025 , pages =. doi:10.1093/nsr/nwaf255 , abstract =

    work page doi:10.1093/nsr/nwaf255 2025

  6. [6]

    and Glicksberg, Benjamin S

    Huang, Kexin and Chandak, Payal and Wang, Qianwen and Havaldar, Shreyas and Vaid, Akhil and Leskovec, Jure and Nadkarni, Girish N. and Glicksberg, Benjamin S. and Gehlenborg, Nils and Zitnik, Marinka , month = dec, year =. A foundation model for clinician-centered drug repurposing , volume =. Nature Medicine , publisher =. doi:10.1038/s41591-024-03233-x ,...

    work page doi:10.1038/s41591-024-03233-x

  7. [7]

    Improving

    Song, Tengwei and Yin, Long and Han, Zhen and Xu, Zhiqiang , month = jan, year =. Improving. doi:10.1145/3770854.3780178 , abstract =

    work page doi:10.1145/3770854.3780178

  8. [8]

    Proceedings of the AAAI Conference on Artificial Intelligence , author =

    Knowledge. Proceedings of the AAAI Conference on Artificial Intelligence , author =. 2025 , pages =. doi:10.1609/aaai.v39i28.35218 , abstract =

    work page doi:10.1609/aaai.v39i28.35218 2025

  9. [9]

    ACS Synthetic Biology , author =

    Simulated. ACS Synthetic Biology , author =. 2023 , pages =. doi:10.1021/acssynbio.3c00186 , abstract =

    work page doi:10.1021/acssynbio.3c00186 2023

  10. [10]

    Monteiro, Lummy M. O. and Chowdhury, Niaz B. and Oostrom, Marjolein T. and McDermott, Jason E. and Stratton, Kelly G. and Choudhury, Sutanay and Bardhan, Jaydeep P. , month = apr, year =. doi:10.64898/2026.04.14.718256 , abstract =

    work page doi:10.64898/2026.04.14.718256 2026

  11. [11]

    A machine learning

    Radivojević, Tijana and Costello, Zak and Workman, Kenneth and Garcia Martin, Hector , month = sep, year =. A machine learning. Nature Communications , publisher =. doi:10.1038/s41467-020-18008-4 , abstract =

    work page doi:10.1038/s41467-020-18008-4

  12. [12]

    Machine learning for metabolic pathway optimization:

    Cheng, Yang and Bi, Xinyu and Xu, Yameng and Liu, Yanfeng and Li, Jianghua and Du, Guocheng and Lv, Xueqin and Liu, Long , month = jan, year =. Machine learning for metabolic pathway optimization:. Computational and Structural Biotechnology Journal , publisher =. doi:10.1016/j.csbj.2023.03.045 , language =

    work page doi:10.1016/j.csbj.2023.03.045 2023

  13. [13]

    Cell , author =

    How to build the virtual cell with artificial intelligence:. Cell , author =. 2024 , keywords =. doi:10.1016/j.cell.2024.11.015 , abstract =

    work page doi:10.1016/j.cell.2024.11.015 2024

  14. [14]

    Adduri, Abhinav K. and Gautam, Dhruv and Bevilacqua, Beatrice and Imran, Alishba and Shah, Rohan and Naghipourfar, Mohsen and Teyssier, Noam and Ilango, Rajesh and Nagaraj, Sanjay and Dong, Mingze and Ricci-Tam, Chiara and Carpenter, Christopher and Subramanyam, Vishvak and Winters, Aidan and Tirukkovular, Sravya and Sullivan, Jeremy and Plosky, Brian S. ...

    work page doi:10.1101/2025.06.26.661135 2025

  15. [15]

    Xin, Kexuan and Wang, Qingyun and Chen, Junyu and Yu, Pengfei and Zhao, Huimin and Ji, Heng , month = oct, year =. Gene-. doi:10.48550/arXiv.2410.18475 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2410.18475

  16. [16]

    and Blake, Judith A

    Ashburner, Michael and Ball, Catherine A. and Blake, Judith A. and Botstein, David and Butler, Heather and Cherry, J. Michael and Davis, Allan P. and Dolinski, Kara and Dwight, Selina S. and Eppig, Janan T. and Harris, Midori A. and Hill, David P. and Issel-Tarver, Laurie and Kasarskis, Andrew and Lewis, Suzanna and Matese, John C. and Richardson, Joel E....

    work page doi:10.1038/75556

  17. [17]

    Building a knowledge graph to enable precision medicine , volume =

    Chandak, Payal and Huang, Kexin and Zitnik, Marinka , month = feb, year =. Building a knowledge graph to enable precision medicine , volume =. Scientific Data , publisher =. doi:10.1038/s41597-023-01960-3 , abstract =

    work page doi:10.1038/s41597-023-01960-3

  18. [18]

    Metabolic Engineering , author =

    Integrated knowledge mining, genome-scale modeling, and machine learning for predicting. Metabolic Engineering , author =. 2021 , keywords =. doi:10.1016/j.ymben.2021.07.003 , abstract =

    work page doi:10.1016/j.ymben.2021.07.003 2021

  19. [19]

    Bioinformatics , author =

    Using graph neural networks for site-of-metabolism prediction and its applications to ranking promiscuous enzymatic products , volume =. Bioinformatics , author =. 2023 , pages =. doi:10.1093/bioinformatics/btad089 , abstract =

    work page doi:10.1093/bioinformatics/btad089 2023

  20. [20]

    Dwivedi, Vijay Prakash and Luu, Anh Tuan and Laurent, Thomas and Bengio, Yoshua and Bresson, Xavier , month = feb, year =. Graph. doi:10.48550/arXiv.2110.07875 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2110.07875

  21. [21]

    , month = mar, year =

    Hasibi, Ramin and Michoel, Tom and Oyarzún, Diego A. , month = mar, year =. Integration of graph neural networks and genome-scale metabolic models for predicting gene essentiality , volume =. npj Systems Biology and Applications , publisher =. doi:10.1038/s41540-024-00348-2 , abstract =

    work page doi:10.1038/s41540-024-00348-2

  22. [22]

    and Riley, Patrick F

    Gilmer, Justin and Schoenholz, Samuel S. and Riley, Patrick F. and Vinyals, Oriol and Dahl, George E. , month = jul, year =. Neural. Proceedings of the 34th

    work page

  23. [23]

    2024 , pages =

    Bioinformatics , author =. 2024 , pages =. doi:10.1093/bioinformatics/btae349 , abstract =

    work page doi:10.1093/bioinformatics/btae349 2024

  24. [24]

    Systematic integration of biomedical knowledge prioritizes drugs for repurposing , volume =

    Himmelstein, Daniel Scott and Lizee, Antoine and Hessler, Christine and Brueggeman, Leo and Chen, Sabrina L and Hadley, Dexter and Green, Ari and Khankhanian, Pouya and Baranzini, Sergio E , editor =. Systematic integration of biomedical knowledge prioritizes drugs for repurposing , volume =. eLife , publisher =. 2017 , keywords =. doi:10.7554/eLife.26726...

    work page doi:10.7554/elife.26726 2017

  25. [25]

    Strategies for Pre-training Graph Neural Networks

    Hu, Weihua and Liu, Bowen and Gomes, Joseph and Zitnik, Marinka and Liang, Percy and Pande, Vijay and Leskovec, Jure , month = feb, year =. Strategies for. doi:10.48550/arXiv.1905.12265 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1905.12265 1905

  26. [26]

    , month = jan, year =

    Oyetunde, Tolutola and Liu, Di and Martin, Hector Garcia and Tang, Yinjie J. , month = jan, year =. Machine learning framework for assessment of microbial factory performance , volume =. PLOS ONE , publisher =. doi:10.1371/journal.pone.0210558 , abstract =

    work page doi:10.1371/journal.pone.0210558

  27. [27]

    Kendall, Alex and Gal, Yarin and Cipolla, Roberto , month = apr, year =. Multi-. doi:10.48550/arXiv.1705.07115 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1705.07115

  28. [28]

    Single-cell biological network inference using a heterogeneous graph transformer , volume =

    Ma, Anjun and Wang, Xiaoying and Li, Jingxian and Wang, Cankun and Xiao, Tong and Liu, Yuntao and Cheng, Hao and Wang, Juexin and Li, Yang and Chang, Yuzhou and Li, Jinpu and Wang, Duolin and Jiang, Yuexu and Su, Li and Xin, Gang and Gu, Shaopeng and Li, Zihai and Liu, Bingqiang and Xu, Dong and Ma, Qin , month = feb, year =. Single-cell biological networ...

    work page doi:10.1038/s41467-023-36559-0

  29. [29]

    Sign and Basis Invariant Networks for Spectral Graph Representation Learning

    Lim, Derek and Robinson, Joshua and Zhao, Lingxiao and Smidt, Tess and Sra, Suvrit and Maron, Haggai and Jegelka, Stefanie , month = sep, year =. Sign and. doi:10.48550/arXiv.2202.13013 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2202.13013

  30. [30]

    URL https://www.science.org/doi/abs/10.1126/science.ade2574

    Lin, Zeming and Akin, Halil and Rao, Roshan and Hie, Brian and Zhu, Zhongkai and Lu, Wenting and Smetanin, Nikita and Verkuil, Robert and Kabeli, Ori and Shmueli, Yaniv and dos Santos Costa, Allan and Fazel-Zarandi, Maryam and Sercu, Tom and Candido, Salvatore and Rives, Alexander , month = mar, year =. Evolutionary-scale prediction of atomic-level protei...

    work page doi:10.1126/science.ade2574

  31. [31]

    Wang, Zehong and Liu, Zheyuan and Ma, Tianyi and Li, Jiazheng and Zhang, Zheyuan and Fu, Xingbo and Li, Yiyang and Yuan, Zhengqing and Song, Wei and Ma, Yijun and Zeng, Qingkai and Chen, Xiusi and Zhao, Jianan and Li, Jundong and Jiang, Meng and Lio, Pietro and Chawla, Nitesh and Zhang, Chuxu and Ye, Yanfang , month = may, year =. Graph. doi:10.48550/arXi...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2505.15116

  32. [32]

    Lobentanzer, Sebastian and Aloy, Patrick and Baumbach, Jan and Bohar, Balazs and Charoentong, Pornpimol and Danhauser, Katharina and Doğan, Tunca and Dreo, Johann and Dunham, Ian and Fernandez-Torras, Adrià and Gyori, Benjamin M. and Hartung, Michael and Hoyt, Charles Tapley and Klein, Christoph and Korcsmaros, Tamas and Maier, Andreas and Mann, Matthias ...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2212.13543

  33. [33]

    Position: Graph Foundation Models are Already Here

    Mao, Haitao and Chen, Zhikai and Tang, Wenzhuo and Zhao, Jianan and Ma, Yao and Zhao, Tong and Shah, Neil and Galkin, Mikhail and Tang, Jiliang , month = may, year =. Position:. doi:10.48550/arXiv.2402.02216 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2402.02216

  34. [34]

    2024 , keywords =

    Molecular Therapy Nucleic Acids , author =. 2024 , keywords =. doi:10.1016/j.omtn.2024.102139 , abstract =

    work page doi:10.1016/j.omtn.2024.102139 2024

  35. [35]

    Nucleic Acids Research , volume =

    Moretti, Sébastien and Tran, Van Du T and Mehl, Florence and Ibberson, Mark and Pagni, Marco , title =. Nucleic Acids Research , volume =. 2021 , month =. doi:10.1093/nar/gkaa992 , url =

    work page doi:10.1093/nar/gkaa992 2021

  36. [36]

    and Butte, Atul J

    Nelson, Charlotte A. and Butte, Atul J. and Baranzini, Sergio E. , month = jul, year =. Integrating biomedical research and electronic health records to create knowledge-based biologically meaningful machine-readable embeddings , volume =. Nature Communications , publisher =. doi:10.1038/s41467-019-11069-0 , abstract =

    work page doi:10.1038/s41467-019-11069-0

  37. [37]

    arXiv.org , author =

    Large-. arXiv.org , author =. 2021 , file =

    work page 2021

  38. [38]

    Representation

    Xu, Keyulu and Li, Chengtao and Tian, Yonglong and Sonobe, Tomohiro and Kawarabayashi, Ken-ichi and Jegelka, Stefanie , month = jul, year =. Representation. Proceedings of the 35th

    work page

  39. [39]

    Durrant and Brian Kang and Dhruva Katrekar and David B

    Eric Nguyen and Michael Poli and Matthew G. Durrant and Brian Kang and Dhruva Katrekar and David B. Li and Liam J. Bartie and Armin W. Thomas and Samuel H. King and Garyk Brixi and Jeremy Sullivan and Madelena Y. Ng and Ashley Lewis and Aaron Lou and Stefano Ermon and Stephen A. Baccus and Tina Hernandez-Boussard and Christopher Ré and Patrick D. Hsu and ...

    work page 2024

  40. [40]

    and Ku, Jerome and Poli, Michael and Brockman, Greg and Chang, Daniel and Gonzalez, Gabriel A

    Brixi, Garyk and Durrant, Matthew G. and Ku, Jerome and Poli, Michael and Brockman, Greg and Chang, Daniel and Gonzalez, Gabriel A. and King, Samuel H. and Li, David B. and Merchant, Aditi T. and Naghipourfar, Mohsen and Nguyen, Eric and Ricci-Tam, Chiara and Romero, David W. and Sun, Gwanggyu and Taghibakshi, Ali and Vorontsov, Anton and Yang, Brandon an...

    work page doi:10.1101/2025.02.18.638918 2025

  41. [41]

    and Oktay, Deniz and Lin, Zeming and Verkuil, Robert and Tran, Vincent Q

    Hayes, Thomas and Rao, Roshan and Akin, Halil and Sofroniew, Nicholas J. and Oktay, Deniz and Lin, Zeming and Verkuil, Robert and Tran, Vincent Q. and Deaton, Jonathan and Wiggert, Marius and Badkundri, Rohil and Shafkat, Irhum and Gong, Jun and Derry, Alexander and Molina, Raul S. and Thomas, Neil and Khan, Yousuf A. and Mishra, Chetan and Kim, Carolyn a...

    work page doi:10.1126/science.ads0018

  42. [42]

    doi:10.1101/2023.10.01.560349 , abstract =

    Su, Jin and Han, Chenchen and Zhou, Yuyang and Shan, Junjie and Zhou, Xibin and Yuan, Fajie , month = apr, year =. doi:10.1101/2023.10.01.560349 , abstract =

    work page doi:10.1101/2023.10.01.560349 2023

  43. [43]

    and Bambrick, Joshua and Bodenstein, Sebastian W

    Abramson, Josh and Adler, Jonas and Dunger, Jack and Evans, Richard and Green, Tim and Pritzel, Alexander and Ronneberger, Olaf and Willmore, Lindsay and Ballard, Andrew J. and Bambrick, Joshua and Bodenstein, Sebastian W. and Evans, David A. and Hung, Chia-Chun and O’Neill, Michael and Reiman, David and Tunyasuvunakool, Kathryn and Wu, Zachary and Žemgul...

    work page doi:10.1038/s41586-024-07487-w

  44. [44]

    Nature Methods , publisher =

    Cui, Haotian and Wang, Chloe and Maan, Hassaan and Pang, Kuan and Luo, Fengning and Duan, Nan and Wang, Bo , month = aug, year =. Nature Methods , publisher =. doi:10.1038/s41592-024-02201-0 , abstract =

    work page doi:10.1038/s41592-024-02201-0

  45. [45]

    Large-scale foundation model on single-cell transcriptomics , volume =

    Hao, Minsheng and Gong, Jing and Zeng, Xin and Liu, Chiming and Guo, Yucheng and Cheng, Xingyi and Wang, Taifeng and Ma, Jianzhu and Zhang, Xuegong and Song, Le , month = aug, year =. Large-scale foundation model on single-cell transcriptomics , volume =. Nature Methods , publisher =. doi:10.1038/s41592-024-02305-7 , abstract =

    work page doi:10.1038/s41592-024-02305-7

  46. [46]

    Flow matching meets biology and life science: a survey , volume =

    Li, Zihao and Zeng, Zhichen and Lin, Xiao and Fang, Feihao and Qu, Yanru and Xu, Zhe and Liu, Zhining and Ning, Xuying and Wei, Tianxin and Liu, Ge and Tong, Hanghang and He, Jingrui , month = jan, year =. Flow matching meets biology and life science: a survey , volume =. npj Artificial Intelligence , publisher =. doi:10.1038/s44387-025-00066-y , abstract =

    work page doi:10.1038/s44387-025-00066-y

  47. [47]

    and Kim, Donghyo and Kalvet, Indrek and Kipnis, Yakov and Coventry, Brian and Altae-Tran, Han Raut and Bauer, Magnus S

    Ahern, Woody and Yim, Jason and Tischer, Doug and Salike, Saman and Woodbury, Seth M. and Kim, Donghyo and Kalvet, Indrek and Kipnis, Yakov and Coventry, Brian and Altae-Tran, Han Raut and Bauer, Magnus S. and Barzilay, Regina and Jaakkola, Tommi S. and Krishna, Rohith and Baker, David , month = jan, year =. Atom-level enzyme active site scaffolding using...

    work page doi:10.1038/s41592-025-02975-x

  48. [48]

    Bran, Andres and Cox, Sam and Schilter, Oliver and Baldassari, Carlo and White, Andrew D

    M. Bran, Andres and Cox, Sam and Schilter, Oliver and Baldassari, Carlo and White, Andrew D. and Schwaller, Philippe , month = may, year =. Augmenting large language models with chemistry tools , volume =. Nature Machine Intelligence , publisher =. doi:10.1038/s42256-024-00832-8 , abstract =

    work page doi:10.1038/s42256-024-00832-8

  49. [49]

    and MacKnight, Robert and Kline, Ben and Gomes, Gabe , month = dec, year =

    Boiko, Daniil A. and MacKnight, Robert and Kline, Ben and Gomes, Gabe , month = dec, year =. Autonomous chemical research with large language models , volume =. Nature , publisher =. doi:10.1038/s41586-023-06792-0 , abstract =

    work page doi:10.1038/s41586-023-06792-0

  50. [50]

    Proceedings of the 2023

    O'Donoghue, Odhran and Shtedritski, Aleksandar and Ginger, John and Abboud, Ralph and Ghareeb, Ali and Rodriques, Samuel , editor =. Proceedings of the 2023. 2023 , pages =. doi:10.18653/v1/2023.emnlp-main.162 , abstract =

    work page doi:10.18653/v1/2023.emnlp-main.162 2023

  51. [51]

    and Lee, Andrew H

    Roohani, Yusuf H. and Lee, Andrew H. and Huang, Qian and Vora, Jian and Steinhart, Zachary and Huang, Kexin and Marson, Alexander and Liang, Percy and Leskovec, Jure , month = oct, year =

    work page

  52. [52]

    and Leskovec, Jure and Coley, Connor W

    Huang, Kexin and Fu, Tianfan and Gao, Wenhao and Zhao, Yue and Roohani, Yusuf H. and Leskovec, Jure and Coley, Connor W. and Xiao, Cao and Sun, Jimeng and Zitnik, Marinka , month = jun, year =. Therapeutics

    work page

  53. [53]

    A database of over 15.000 strain design publications reveals a conserved set of metabolic engineering targets across microbial hosts and products , copyright =

    Márquez-Zavala, Elisa and Bartolomeu, Francesca Di and Machado, Daniel , month = dec, year =. A database of over 15.000 strain design publications reveals a conserved set of metabolic engineering targets across microbial hosts and products , copyright =. doi:10.64898/2025.12.15.694291 , abstract =

    work page doi:10.64898/2025.12.15.694291 2025

  54. [54]

    Biotechnology and Bioengineering , author =

    Optknock:. Biotechnology and Bioengineering , author =. 2003 , keywords =. doi:10.1002/bit.10803 , abstract =

    work page doi:10.1002/bit.10803 2003

  55. [55]

    2022 , pages =

    Bioinformatics , author =. 2022 , pages =. doi:10.1093/bioinformatics/btac632 , abstract =

    work page doi:10.1093/bioinformatics/btac632 2022

  56. [56]

    Akiba, Takuya and Sano, Shotaro and Yanase, Toshihiko and Ohta, Takeru and Koyama, Masanori , month = jul, year =. Optuna:. Proceedings of the 25th. doi:10.1145/3292500.3330701 , abstract =

    work page doi:10.1145/3292500.3330701

  57. [57]

    Fey, Matthias and Lenssen, Jan Eric , month = apr, year =. Fast. doi:10.48550/arXiv.1903.02428 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1903.02428 1903

  58. [58]

    2000 , pages =

    Nucleic Acids Research , author =. 2000 , pages =. doi:10.1093/nar/28.1.27 , abstract =

    work page doi:10.1093/nar/28.1.27 2000

  59. [59]

    2015 , pages =

    Bioinformatics , author =. 2015 , pages =. doi:10.1093/bioinformatics/btu739 , abstract =

    work page doi:10.1093/bioinformatics/btu739 2015

  60. [60]

    2023 , pages =

    Nucleic Acids Research , author =. 2023 , pages =. doi:10.1093/nar/gkac1052 , abstract =

    work page doi:10.1093/nar/gkac1052 2023

  61. [61]

    2023 , pages =

    Nucleic Acids Research , author =. 2023 , pages =. doi:10.1093/nar/gkac993 , abstract =

    work page doi:10.1093/nar/gkac993 2023

  62. [62]

    Nucleic Acids Research , author =

    Gene: a gene-centered information resource at. Nucleic Acids Research , author =. 2015 , keywords =. doi:10.1093/nar/gku1055 , url =

    work page doi:10.1093/nar/gku1055 2015

  63. [63]

    2020 , pages =

    Database , author =. 2020 , pages =. doi:10.1093/database/baaa062 , abstract =

    work page doi:10.1093/database/baaa062 2020

  64. [64]

    and Deng, Kai and Deutsch, Samuel and Baidoo, Edward E

    Opgenorth, Paul and Costello, Zak and Okada, Takuya and Goyal, Garima and Chen, Yan and Gin, Jennifer and Benites, Veronica and de Raad, Markus and Northen, Trent R. and Deng, Kai and Deutsch, Samuel and Baidoo, Edward E. K. and Petzold, Christopher J. and Hillson, Nathan J. and Garcia Martin, Hector and Beller, Harry R. , title =. ACS Synthetic Biology ,...

    work page 2019

  65. [65]

    Nature Communications , author =

    A genome-scale. Nature Communications , author =. 2016 , pages =. doi:10.1038/ncomms13806 , url =

    work page doi:10.1038/ncomms13806 2016