REVIEW 3 major objections 6 minor 19 references
DoGMA: A Central-Dogma-Guided Foundation Model for Multi-Omics Alignment and Multi-Task Learning in Oncology
T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash
Pith's one-line read DoGMA argues that making multi-omics attention follow the central dogma’s DNA-to-RNA-to-protein direction improves cancer representations and transfer.
desk verdict A genuinely new directed-attention mechanism for multi-omics with consistent results; the missing cancer-label ablation and private survival cohort weaken the attribution, but the core idea holds up. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the omics-specific residual Central-Dogma attention bias: for each head, logits get an additive term $B_{\text{dogma}}$ that is a fixed negative constant $A$ on weak-feedback (reverse-direction) edges plus a zero-initialized learnable residual $R$ per omics pair, with a binary topology matrix $T$ marking which source-to-target edges follow the dogma. A second piece is the masked hierarchical reconstruction pretraining: a VQ-VAE tokenizer maps each modality to discrete tokens, and the backbone is trained to reconstruct masked tokens within a modality and to reconstruct each modality's latent from the fused representation, so the model must actually use cross-omics routes to recover missing molecular states. A third piece is the modality-specific Mixture-of-Experts fusion that lets each omics layer be transformed by its own expert with learned routing weights. These pieces together give the model a directional prior, a training pressure to use it, and a mechanism to adapt it per pair and per sample.
What would settle it
An ablation that removes only the cancer-type semantic label from the supervised contrastive loss, keeping the directional attention and all reconstruction terms, should be run and compared with full DoGMA on METABRIC test accuracy and the Dogma Directionality Index. If the label-free variant loses most of the 0.955 test accuracy or the DDI gap between DoGMA and standard attention, then the label, rather than the directional prior, is carrying the reported gains.
Extended reading notes
Core claim
On its own terms, DoGMA establishes that the central dogma can be operationalized as a structural prior for attention: each omics pair gets a directed communication edge, forward (DNA to RNA to protein) edges are favored by a shared negative bias, reverse edges are penalized, and a zero-initialized learnable residual keeps pair-specific flexibility. The paper shows that this prior, when combined with reconstruction objectives that force cross-omics inference, yields representations that separate cancer types more clearly (silhouette 0.628 versus 0.368 for the strongest baseline), reconstruct gene expression from non-expression omics with higher correlation, infer driver mutation status with higher AUC, and transfer to an external breast-cancer cohort with higher average accuracy than bidirectional baselines. The paper also reports that attention-flow directionality, measured by a Dogma Directionality Index, tracks downstream performance, and that inference-time knockout of forward edges is the most damaging intervention while restoring forward edges is the most effective rescue. The intended upshot is that the performance gain is not capacity but biologically consistent information-flow structure learned through pretraining.
Load-bearing premise
The load-bearing premise is that the performance gains are caused by the central-dogma directionality prior and the reconstruction objectives, not by the 32-category cancer-type label that is included in the supervised contrastive loss during pretraining; the paper provides no ablation that removes only that label.
Editorial extensions
If this is right
- If the claim is right, future multi-omics architectures should replace symmetric cross-modal attention with topology-aware attention that encodes known biological directionality, rather than leaving interactions fully unconstrained.
- Reconstruction-based pretraining becomes a necessary companion to the prior: the ablations show that removing cross-omics reconstruction is the single largest drop in external-cohort accuracy, so the prior only pays off when training forces the model to use it.
- Missing modality settings benefit directly: a missing omics layer is treated as an unobserved molecular state that can be inferred through forward routes, which explains robust performance under incomplete assay coverage.
- Inference-time edge interventions provide a testable signature: if a model has internalized the prior, knocking out forward edges should degrade prediction more than knocking out reverse or random edges, which is what DoGMA shows.
- The Dogma Directionality Index offers a model-level diagnostic that correlates with downstream performance and could be used to audit whether a multi-omics model is relying on biologically plausible information flow.
Reading between the lines
- The paper never ablates the 32-category cancer-type label used in the supervised contrastive loss, so the cleanest test of its attribution would be an ablation that removes only that label; if the silhouette jump and transfer gains mostly vanish, the label rather than the directional prior would carry the result.
- The same directed-attention-plus-reconstruction recipe could be applied to other biological or physical systems with an ordering constraint, such as signaling cascades, metabolic pathways, or time-ordered multimodal data, where the directionality is known a priori.
- A stronger falsification would come from comparing the Dogma Directionality Index against experimentally validated perturbation effects: if high-DDI models do not predict regulatory responses better than low-DDI models on an independent gene-perturbation benchmark, the directionality metric is not actually tracking biological consistency.
- Because the pretraining cohort is TCGA-only, the promised pan-cancer generalization across platforms would be better supported by a second external cohort beyond METABRIC and the institutional colorectal set, testing whether the transferred representations survive platform shifts elsewhere.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper introduces DoGMA, a Transformer-Mixture-of-Experts foundation model for pan-cancer multi-omics analysis. Its central design idea is to impose a directional attention bias that follows the central dogma of molecular biology (DNA to RNA to protein, with weak feedback), and to pretrain with masked hierarchical reconstruction, cross-omics reconstruction, adversarial alignment, and supervised contrastive learning. Downstream evaluations include pan-cancer representation geometry, cross-cohort breast cancer subtyping from TCGA to METABRIC, metastasis prediction on a TCGA MetaCancer cohort, and survival analysis on an institutional COAD/READ cohort. The paper reports improved silhouette scores, an average METABRIC test accuracy of 0.955, a metastasis F1 of 0.975, and better C-index values, together with ablations and attention-flow analyses, and attributes the gains to a synergy between the directed attention prior and reconstruction-based pretraining.
Significance. If the empirical claims hold, the paper offers a practically useful demonstration that a biologically motivated directional bias can improve transfer in multi-omics foundation models. The evaluation has notable strengths: five-seed downstream runs, case-level splitting in the metastasis cohort, an external METABRIC cohort that was held out from model development, parameter-matched ablation variants, and an attempted mechanistic analysis through DDI and edge interventions. However, the central attribution is underdetermined because pretraining includes a strong cancer-type semantic label whose contribution is not isolated. In addition, two of the three headline transfer results rest on a small test set and a private cohort, limiting independent verification. With the missing ablation and additional statistical evidence, the contribution would be significant for the multi-omics representation learning community.
major comments (3)
- [Eq. (10) and Table 2] The central claim that the gains come from the central-dogma prior is not supported, because the pretraining objective in Eq. (10) includes L_supcon, a supervised contrastive loss using a 32-category cancer-type semantic label (Methods: 'Cancer type was encoded as a 32-category semantic label during representation learning'). A supervised contrastive loss is specifically designed to cluster same-label samples and separate different labels, so the silhouette gain from 0.368 to 0.628 in Figure 2 and part of the downstream transfer gain in Table 1 (average test accuracy 0.955) could be driven by this label rather than by the directed attention prior. The existing ablations do not isolate the label: 'w/o Contra' removes the entire contrastive term, confounding label supervision with the contrastive mechanism, and 'w/o Adv-GRL' retains the label while removing only the adversarial term. The paper needs an ablation that removes only the cancer-type label, for example by replacing L_supcon with an unsupervised or batch-level contrastive objective while keeping all other components unchanged; without such an ablation, the attribution of the representation-geometry and transfer gains to the central-dogma prior is underdetermined.
- [Attention-flow Alignment, Tables 3 and 4] The DDI analysis and counterfactual edge intervention are presented as evidence that DoGMA 'functionally relies' on prior-consistent routes, but the link is weaker than stated. DDI is computed from the same edge taxonomy that defines the directional prior, and the variants in Table 3 differ in ways that directly change how much attention can flow along those edges (standard attention, scratch training, removing reconstruction objectives), so the positive association between DDI and performance is partly by construction and is based on only five points without confidence intervals. The edge-knockout result in Table 4 is suggestive, but applying a hard test-time mask to every attention block can induce a large distribution shift relative to the training-time computation, so the performance drop does not by itself establish learned mechanistic reliance. Please provide control interventions with random edge sets of matched cardinality beyond what is reported, bootstrapped confidence intervals for DDI, and ideally a control model with a shuffled topology.
- [Metastasis Prediction on TCGA; Institutional COAD/READ] Two of the three headline external-transfer results rest on limited or inaccessible test sets. The metastasis test set contains 79 specimens from 64 unique cases (Table 7); Figure 4 reports means over five seeds but no confidence intervals or significance tests against the baselines, and with this sample size the reported F1 gap (0.975 for pretrained DoGMA versus 0.916 for pretrained TMO-Net) may not be stable. The survival experiment uses a private institutional cohort whose data and partition manifests will be released only upon acceptance, so reviewers cannot verify the survival preprocessing, endpoint definitions, or the Cox results. Please report case-level bootstrapped confidence intervals and significance tests for the metastasis metrics, and either release the survival cohort or validate on a public colorectal cancer dataset (for example TCGA-COAD/READ with appropriate splits) to make the survival claim independently checkable.
minor comments (6)
- [Hyperparameter Development] The breast-cancer fine-tuning protocol uses a learning rate of 1e-5 for pretrained DoGMA but 1e-4 for scratch and TMO-Net baselines; this creates a small confound, and the paper should justify or test sensitivity to this choice.
- [Pretraining configuration] The checkpoint interpolation described in the Pretraining Configuration is not fully specified: the interpolation coefficient is not reported, so the exact initialization used for downstream adaptation is not reproducible.
- [Table 2 caption] The 'Breast subtype' columns in Table 2 appear to report averages over the four breast-cancer tasks, but the caption does not state this; please clarify to match the 'Avg.' convention used in Table 1.
- [Figure 2] The text does not state which split (training, validation, or test) is used for the t-SNE visualization and silhouette scores in Figure 2; please specify the split in the caption.
- [Table 5 and causal-chain analysis] The in-silico intervention results in Table 5 are described as hypothesis-generating in the appendix, but the main text refers to 'causal-chain analysis'; please include an explicit statement in the main text that these results are model-generated hypotheses, not experimentally validated causal mechanisms.
- [Figure 5] The survival C-index results in Figure 5 are shown as distributions, but no confidence intervals or pairwise significance tests are reported; adding these would strengthen the survival claim.
Circularity Check
No significant circularity: DoGMA's central transfer results are empirical comparisons against external cohorts and baselines; the cancer-type label in pretraining is a confound for attribution, not a circular derivation.
full rationale
The paper's central performance claims—METABRIC cross-cohort breast-cancer subtyping, TCGA metastasis prediction, and COAD/READ survival analysis—are empirical evaluations against independent held-out cohorts and external baseline models. None of the reported accuracies, AUCs, or C-indices is obtained by construction from a fitted parameter or from the pretraining objective. The pretraining loss in Eq. (10) does include a supervised contrastive term with a 32-category cancer-type label, and Figure 2's silhouette improvement is measured on the same cancer-type structure that this label encourages; however, the silhouette metric is not identical to the contrastive loss, and the downstream benchmarks are not optimized by the label. This is a potential confound for attributing gains to the central-dogma attention prior, but it is a missing-control/correctness issue rather than a circular derivation. The DDI analysis is descriptive: it correlates an attention-flow summary with downstream performance across variants; although the attention bias in Eq. (5) shapes DDI by design, the association with external performance is empirical, not tautological. The paper contains no load-bearing self-citation, no imported uniqueness theorem, and no ansatz smuggled via citation. The derivation chain is therefore self-contained with respect to circularity.
Assumptions & free parameters
free parameters (6)
- Reverse-direction penalty A =
-1 or -2 (grid searched)
- Cross-omics reconstruction weight lambda_1 =
Not reported
- Supervised contrastive weight lambda_con =
0.5 or 1.0
- Adversarial alignment weight lambda_adv =
0.05 with linear warmup over 60 epochs
- Masking ratio for hierarchical reconstruction =
0.15
- Checkpoint interpolation coefficient =
Not reported
assumptions (5)
- domain assumption Central dogma directionality (DNA to RNA to protein, weak feedback) is the dominant cross-omics information flow for cancer representation learning.
- domain assumption The omics ordering {methylation, mutation, CNA, expression, protein, clinical} reflects a valid molecular-level ordering for the topology matrix T.
- standard math Masked reconstruction of VQ-VAE latents is a sufficient pretraining objective to induce transferable cross-omics representations.
- domain assumption The 32-category cancer-type label can be used as a semantic label during pretraining without making downstream evaluations leak.
- ad hoc to paper Validation silhouette score is a useful criterion for checkpoint selection.
Cite this review
Pith. "Pith review of DoGMA: A Central-Dogma-Guided Foundation Model for Multi-Omics Alignment and Multi-Task Learning in Oncology." pith.science (2026). https://pith.science/paper/QYX64SEI
@misc{pith2026260808148,
author = {Pith},
title = {Pith review of: DoGMA: A Central-Dogma-Guided Foundation Model for Multi-Omics Alignment and Multi-Task Learning in Oncology},
year = {2026},
howpublished = {\url{https://pith.science/paper/QYX64SEI}},
note = {Machine review of arXiv:2608.08148}
}
read the original abstract
Attention mechanisms have been widely utilized in modern deep learning, and many existing multi-omics models inherit their conventional use to allow unrestricted bidirectional interactions. However, the fundamental logic of life is directional. Existing designs often overlook the directionality suggested by the central dogma, potentially limiting transfer across heterogeneous cancers, downstream tasks, and incomplete modality settings.In this work, we present DoGMA, a central-dogma-guided foundation model for pan-cancer multi-omics analysis, arguing that robust transfer requires representations with domain-specific inductive bias. Concretely, we build it on a Transformer-MoE architecture where directed attention biases inter-omics communication toward central-dogma information flow. We further pretrain our model with masked hierarchical omics reconstruction to guide it toward learning central-dogma-consistent interactions. Across diverse downstream tasks, including cancer representation learning, survival prediction, and metastasis prediction, DoGMA consistently demonstrates strong predictive performance. Ablations and analyses further suggest that the performance gains arise from the synergy between central-dogma-guided directed attention and reconstruction-based pretraining, which together promote more biologically consistent cross-omics information exchange. Overall, DoGMA demonstrates that domain-specific inductive biases can improve the robustness and transferability of multi-omics foundation models, offering new insights into the design of attention mechanisms for multi-omics representation learning.
Figures
Figures from the paper (5 more)
Reference graph
Works this paper leans on
-
[1]
Quantifyingattentionflow in transformers
Abnar,S.;andZuidema,W.2020. Quantifyingattentionflow in transformers. InProceedings of the 58th annual meeting oftheassociationforcomputationallinguistics,4190–4197. Alayrac, J.-B.; Donahue, J.; Luc, P.; Miech, A.; Barr, I.; Hasson, Y.; Lenc, K.; Mensch, A.; Millican, K.; Reynolds, M.; et al
work page 2020
-
[8]
Predicting transcriptional outcomes of novel multigene perturbations with GEARS.Nature Biotechnology, 42(6): 927–935. Rudin,C.2019. Stopexplainingblackboxmachinelearning modelsforhighstakesdecisionsanduseinterpretablemodels instead.Nature machine intelligence, 1(5): 206–215. Schölkopf, B.; Locatello, F.; Bauer, S.; Ke, N. R.; Kalch- brenner,N.;Goyal,A.;an...
work page 2019
-
[13]
models interac- tions among cancer-specific pathway representations. These advances show that biologically informed architectures can substantially improve multi-omics modeling. DoGMA extends this line of research by introducing bio- logical structure directly into communication among omics- level representations. Rather than treating molecular modal- iti...
work page 2024
-
[14]
provides another im- portant precedent by mapping epigenomic, genomic, tran- scriptomic, and proteomic measurements onto multi-level signaling graphs. Its graph architecture propagates informa- tion through promoter-to-protein routes and learns molec- ular structure through masked interaction reconstruction. DoGMA captures a complementary level of biologi...
work page 2021
-
[15]
DoGMA introduces modality-specific expert specializa- tion into multi-omics fusion
demonstrate the effectiveness of conditional expert computation at scale, while multimodal models such asI 2MoE(Xinetal.2025)usespecializedexpertstocapture heterogeneous interactions among modalities. DoGMA introduces modality-specific expert specializa- tion into multi-omics fusion. Following central-dogma- guidedattention,eachmolecularrepresentationispr...
work page 2025
-
[16]
EGFRCopy NumberIGHG1(Liu et al. 2021),IGKC(Liu et al. 2021),IGLC2(Liu et al. 2021),IGHG2(Liu et al. 2021), MT-CO2, MT-ND2, KRT5, MT-CO3, MT-RNR2, MT-ND4 CCND1Copy NumberKRT5(Tsang, Tam, and Oh 2023),KRT6A(Tsang, Tam, and Oh 2023),KRT14(Tsang, Tam, and Oh 2023), SPRR1B, KRT17, KRT16, SPRR1A, ANXA8, LY6D,S100A8(Tsang, Tam, and Oh
work page 2021
-
[17]
2021), RPS4Y1,COL1A2(Joshi et al
FGFR1Copy Number SPINK1,TNC(Joshi et al. 2021), RPS4Y1,COL1A2(Joshi et al. 2021), KRT14, MT-CO2, KRT6A, GPX2,COL1A1(Joshi et al. 2021),POSTN(Joshi et al
work page 2021
-
[18]
Table 5: Biological plausibility of DoGMA-guided in-silico interventions. For each upstream driver node, we perturb the corresponding omics feature and rank downstream RNA or protein targets by the magnitude of their reconstructed response. The table reports the top affected targets and representative literature priors supporting their association with ca...
work page 2023
Show all 19 references
-
[30]
Attention is all you need.Advances in neural information processing systems,
Vaswani,A.;Shazeer,N.;Parmar,N.;Uszkoreit,J.;Jones,L.; Gomez,A.N.;Kaiser,Ł.;andPolosukhin,I.2017. Attention is all you need.Advances in neural information processing systems,
2017
-
[49]
I2moe:Interpretablemultimodalinteraction- awaremixture-of-experts.arXivpreprintarXiv:2505.19190
Xin,J.;Yun,S.;Peng,J.;Choi,I.;Ballard,J.L.;Chen,T.;and Long,Q.2025. I2moe:Interpretablemultimodalinteraction- awaremixture-of-experts.arXivpreprintarXiv:2505.19190. Zeng, Y.; Xie, J.; Wei, Z.; Su, Y.; Shangguan, N.; Yang, S.; Zhang,C.;Li,W.;Zhang,J.;Fang,N.;etal.2024. CellFM:a...
2025 arXiv
-
[149]
MOGONETintegratesmulti-omics data using graph convolutional networks allowing patient classification and biomarker identification.Nature commu- nications, 12(1):
Wang, T.; Shao, W.; Huang, Z.; Tang, H.; Zhang, J.; Ding, Z.;andHuang,K.2021. MOGONETintegratesmulti-omics data using graph convolutional networks allowing patient classification and biomarker identification.Nature commu- nications, 12(1):
2021
-
[447]
Weretainedpatientswithpartiallyobservedmodalitysets rather than restricting the analysis to complete cases
A total of 436 patients had all five molecular modalities. Weretainedpatientswithpartiallyobservedmodalitysets rather than restricting the analysis to complete cases. This protocolallowsthesurvivalexperimenttoreflecttheincom- plete assay coverage encountered in the institution...
2024
-
[2015]
InInternational conference on machine learning, 1180–1189
Unsupervised domain adaptation by backpropagation. InInternational conference on machine learning, 1180–1189. PMLR. Hama, H. A.; Hasan, B. S.; Barzngy, B.; Abdulla, S. S.; Karim,A.Y.;Khailany,R.A.;Miasko,M.;Dabrowski,J.M.; andPucelik,B.2023. KRASandMT-CO1genesincolorec- talcan...
2023
-
[2019]
Ashuach, T.; Gabitto, M
Invariant risk minimization.arXiv preprint arXiv:1907.02893. Ashuach, T.; Gabitto, M. I.; Koodli, R. V.; Saldi, G.-A.; Jor- dan, M. I.; and Yosef, N
1907 arXiv
-
[2021]
Battaglia, P
Beit: Bert pre-training of image transformers.arXiv preprint arXiv:2106.08254. Battaglia, P. W.; Hamrick, J. B.; Bapst, V.; Sanchez- Gonzalez, A.; Zambaldi, V.; Malinowski, M.; Tacchetti, A.; Raposo,D.;Santoro,A.;Faulkner,R.;etal.2018. Relational inductive biases, deep learnin...
2018 arXiv
-
[2022]
Albaradei, S.; Napolitano, F.; Thafar, M
Flamingo: a visual language model for few- shot learning.Advances in neural information processing systems, 35: 23716–23736. Albaradei, S.; Napolitano, F.; Thafar, M. A.; Gojobori, T.; Essack,M.;andGao,X.2021.MetaCancer:Adeeplearning- basedpan-cancermetastasispredictionmodelde...
2021
-
[2023]
Bahdanau,D.;Cho,K.;andBengio,Y.2014.Neuralmachine translation by jointly learning to align and translate.arXiv preprint arXiv:1409.0473
MultiVI: deep generative model for the integration of multimodal data.Nature meth- ods, 20(8): 1222–1231. Bahdanau,D.;Cho,K.;andBengio,Y.2014.Neuralmachine translation by jointly learning to align and translate.arXiv preprint arXiv:1409.0473. Bailey, M. H.; Tokheim, C.; Porta-...
2014 arXiv
-
[2024]
CancerGenomeAtlasResearchNetwork,J.;etal.2013
DeePathNet: A Transformer- Based Deep Learning Model Integrating Multiomic Data with Cancer Pathways.Cancer Research Communications, 4(12): 3151–3164. CancerGenomeAtlasResearchNetwork,J.;etal.2013. The cancergenomeatlaspan-canceranalysisproject.Nat.Genet, 45(10): 1113–1120. Ch...
2013
-
[2025]
Deep learning–driven multi-omics analysis: Enhancing can- cerdiagnosticsandtherapeutics.Briefingsinbioinformatics, 26(4): bbaf440. Zhao, S. G.; Chen, W. S.; Das, R.; Chang, S. L.; Tomlins, S. A.; Chou, J.; Quigley, D. A.; Dang, H. X.; Barnard, T. J.; Mahal,B.A.;etal.2019. Clin...
2019
Reviewed August 12, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.