Pith. sign in

REVIEW 4 major objections 4 minor 44 references

Deep Learning and Explainable AI: New Pathways to Genetic Insights

T0 review · 4 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that the limitations of input-based and model-based interpretability in genomics can be derived mathematically, not merely observed.

desk verdict The survey of interpretable deep learning in genomics is useful, but the advertised proofs of limitations are not sound—the key proof in §3.2 assumes the inequality it needs to establish. read the letter →

arxiv 2505.09873 v1 pith:LM63RN7D submitted 2025-05-15 q-bio.GN

Chenyu Wang , Chaoying Zuo , Zihan Su , Yuhang Xing , Lu Li , Maojun Wang , Zeyu Zhang This is my paper · ORCID
classification q-bio.GN
keywords explainableAIgenomicsdeeplearninginterpretabilityattentionmechanismtransparentmodelsvanishinggradientmulticollinearitymathematicalderivation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper is a review of interpretable deep learning in genomics that goes further than cataloguing methods: it tries to prove, from the ground up, why several widely used interpretation tools can be misleading. It argues that unconstrained convolutional kernels can grow into distorted position weight matrices, dropout-trained networks can hide feature importance in redundant neurons, deep ReLU networks can suffer vanishing gradients, multicollinear DNA features can make attention weights numerically unstable, and hard-coded biologically transparent models can lose predictive performance. If the derivations hold, the limitations of input-based and model-based interpretability in genomics stop being empirical folklore and become checkable mathematical facts that a researcher can test on their own model. The review also gives a practical map of current methods and future directions, but the theoretical claims are the pith.

What carries the argument

The machinery is a set of self-contained derivations attached to each interpretability class. The convolutional-kernel argument expands the update $\partial L/\partial w_{11}=(\sigma(z)-y)(\alpha_1 s_{11}+\cdots+\alpha_L s_{L1})$ to show unconstrained weights can grow, then applies the softmax transform $w'_{11}=e^{w_{11}}/(e^{w_{11}}+e^{w_{12}}+e^{w_{13}}+e^{w_{14}})$ followed by $\log(w'_{ij}+\epsilon)$ to show how extreme entries distort the position weight matrix. The dropout argument treats neurons as parameter directions in a high-dimensional manifold and identifies redundant directions through commuting covariant derivatives, giving a locally flat Riemann tensor $R(v_i,v_j,v_i,v_j)\approx 0$; removal of such a neuron changes the output by $O(\epsilon^2)$. The vanishing-gradient argument multiplies per-layer factors $\mathrm{ReLU}'(W^{(l)}Z^{(l-1)}+b^{(l)})W^{(l)}$ and notes the product can go to zero. The attention argument proves Lemma 1 ($\|A\|_2=\sqrt{\lambda_{\max}(A^TA)}$) and then Theorem 2: multicollinearity forces $\lambda_{\min}=0$, so the condition number $\kappa(A)=\|A\|_2\|A^{-1}\|_2\to\infty$. The transparent-model argument compares $L_{\text{hard}}$ and $L_{\text{free}}$ term by term, reducing the comparison to the pointwise inequality $p^{\text{hard}}_i\le p^{\text{free}}_i$.

What would settle it

Train the same architecture twice on the same genomic dataset, once with a hard-coded biological prior and once unconstrained, and record per-sample predicted probabilities; if any sample has $p^{\text{hard}}_i > p^{\text{free}}_i$ while both models are near convergence, the assumed inequality fails and the claimed loss ordering can reverse.

Watch

Extended reading notes

Core claim

The paper's central claim is that the known weaknesses of interpretable methods in genomic deep learning are not just anecdotal: they follow from explicit mathematical statements about the trained model. Concretely, it asserts that (1) unconstrained gradient-descent updates on convolutional kernels can drive individual weights so large that the softmax-to-log conversion used in motif visualization distorts the importance of sequence positions; (2) dropout's random deactivation forces redundant neurons, so removing one neuron changes output only to second order and perturbation-based attribution understates its importance; (3) the product of layer-wise gradients in ReLU networks can vanish, making gradient-based saliency unreliable; (4) when key vectors in attention are multicollinear, the covariance matrix $K^TK$ is singular and the condition number of the attention matrix tends to infinity, implying numerical instability; and (5) a hard-coded transparent model has loss no less than an unconstrained model when per-sample predicted probabilities satisfy $p^{\text{hard}}_i\le p^{\text{free}}_i$, so biological hard-coding can degrade performance.

Load-bearing premise

The load-bearing premise is that, sample by sample, the hard-coded transparent model's predicted probability never exceeds the unconstrained model's; that inequality is asserted rather than proven, and the conclusion that hard-coded models perform worse would not follow if any sample violates it.

Editorial extensions

If this is right

  • Convolutional kernel visualizations should not be read as motifs unless the training imposed constraints on kernel weights; without constraints, near-zero or near-one softmax entries can dominate the log-transformed PWM.
  • In networks trained with dropout, small output changes after deleting a neuron do not establish that the neuron is unimportant, so perturbation-based attribution maps need redundancy correction to be trustworthy.
  • For deep ReLU models, zero or tiny input gradients do not mean the input is irrelevant; gradient-based saliency should be paired with path-integral or reference-based methods.
  • Attention heatmaps from DNA sequences with correlated positions are numerically unstable; the condition number of the key-feature covariance should be reported before interpreting attention weights biologically.
  • Transparent models that hard-code biological prior interactions can be expected to show higher loss than unconstrained models unless the prior constraints are regularized gently, giving a concrete trade-off between interpretability and predictive accuracy.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The condition-number criterion suggests a practical pre-check: compute the smallest eigenvalue of the key-feature covariance before trusting attention; if $\lambda_{\min}$ is near machine precision, attention weights should be treated as arbitrary.
  • The loss-difference proof implies a testable scaling law: the performance gap of a hard-coded transparent model should grow with $\sum_i m_i\log(p^{\text{free}}_i/p^{\text{hard}}_i)$, so measuring per-sample probability gaps on a benchmark would quantify how much biology has to be lost to gain transparency.
  • The same loss-comparison template could be extended to other transparency constraints, such as sparse connectivity, modular layers, or fixed motif detectors, by swapping the prior-constraint term in $L_{\text{hard}}$.
  • The manifold argument yields a direct experiment: train two networks with and without dropout, delete matched neurons, and measure output change; the dropout-trained network should show smaller deletion effects in directions where the local curvature is flat.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper reviews interpretable deep learning methods in genomics, categorizing them into input-based approaches (convolutional kernel visualization, gradient-based methods, perturbation-based methods) and model-based approaches (attention mechanisms and transparent models based on biological prior knowledge). It surveys applications of these methods to 3D genomics and regulatory genomics, and then attempts to provide formal mathematical derivations of five limitations: convolutional weight scaling instability, dropout-induced neuronal redundancy, vanishing gradients, attention weight instability under multicollinearity, and performance loss in hard-coded transparent models. The paper's advertised contribution is to rigorously reveal the inherent limitations of selected interpretable approaches through mathematical derivation.

Significance. The survey component is useful: the categorization is clear, Table 1 collects relevant tools with URLs and biological applications, and the discussion of concrete use cases (Basset, Enformer, DCell, GenNet) is informative. Some mathematical material, such as Lemma 1 and the standard chain-rule derivation of gradient decay, is correct. However, the paper's central claim of rigorous derivation is not supported: the transparent-model proof assumes its conclusion, the dropout-redundancy 'manifold' argument is not a valid theorem, and the attention condition-number result is applied to a singular matrix whose inverse does not exist. If the theoretical claims were removed or explicitly reframed as heuristic observations, the manuscript could serve as a useful review, but as it stands it does not deliver the formal results it advertises.

major comments (4)
  1. [§3.2, Eqs. (23)-(24)] The proof that hard-coded transparent models underperform unconstrained models assumes the conclusion. The pointwise inequality p_hard_i ≤ p_free_i is asserted without derivation after Eq. (24), yet this inequality is exactly the performance-degradation claim being proved. In addition, p_hard_i is defined as f(Wprior, W_adapt, m_i), making the predicted probability a function of the true label, which is not a valid model specification. The regularizer difference term λ||Wprior − Wref_prior||² + γ||Wadapt||² − β||Wfree||² ≈ 0 is also asserted without justification of the relative magnitudes of the coefficients or norms. The advertised rigorous derivation of transparency-induced performance loss therefore collapses at this step.
  2. [§3.1, 'The redundancy of neurons', Eqs. (7)-(9)] The dropout-redundancy argument is not a theorem. Eq. (7) already assumes that perturbations in directions vi and vj yield approximately equal outputs for all inputs, which is the definition of redundancy being used to conclude that removing one neuron changes the output little. The introduction of covariant derivatives and the Riemann curvature tensor is not load-bearing: fθ is a scalar-valued function, so the inner product ⟨∇vi∇vj fθ − ∇vj∇vi fθ, fθ⟩ and 'components of the Riemann curvature tensor' are not defined for scalar fields. Moreover, removing a neuron is a discrete structural change, not a small parameter perturbation αvi, so the Taylor expansion in α does not apply. The conclusion Δf ≈ 0 therefore does not follow from the stated assumptions.
  3. [§3.2, 'Instability of in the estimation of the weights', Theorem 2 and Eq. (22)] Theorem 2 applies the formula ||A^{-1}||₂ = 1/√λmin to a matrix that is singular by construction. Since multicollinearity is defined by r(K) < n, the matrices K and C = K^T K have zero eigenvalues, so A^{-1} does not exist and the standard condition number κ(A) = ||A||₂ ||A^{-1}||₂ is undefined for the matrix at hand. The conclusion that the condition number 'tends to infinity' confuses the unbounded behavior of nearly singular matrices with the undefined quantity for exactly singular matrices. The argument could be repaired by considering κ(A + δI) with δ → 0 or by using the pseudo-inverse condition number, but this is not what is written.
  4. [§3.1, 'The unconstrained learning of weights', Eqs. (1)-(5)] The claim that unconstrained weight optimization 'provably causes' scaling instability is not established. The derivation shows only that if σ(z) − y < 0 and the subsequent factor α1s11 + ... + αLsL1 is positive, then w11 increases; it does not show that this condition holds, that the growth is unbounded, or that a constrained variant would avoid the issue. Additionally, Eq. (1) appears to contain a sign error: standard gradient descent is w_ij ← w_ij − η ∂L/∂w_ij, not w_ij ← −w_ij − η ∂L/∂w_ij. As written, the section is an informal illustration of a known risk, not a rigorous proof of scaling instability.
minor comments (4)
  1. [Throughout] Typographical errors appear throughout: 'DeePLIFT' should be 'DeepLIFT', 'Bernouli' should be 'Bernoulli', 'Multicolinearity' should be 'multicollinearity', and 'whichϵ is always determined' is missing a space before the epsilon.
  2. [References, [33]] Reference [33] is mis-titled: the cited arXiv:1611.02639 paper is 'Axiomatic Attribution for Deep Networks', not 'Gradients of counterfactuals'.
  3. [§3.1, 'Vanishing gradient in DNA sequences'] In the final product, the ordering of weight matrices is presented inconsistently: the sentence 'When W(1), W(L−1),...,W(L) are very small' does not match the previous product W^{(L)} W^{(L−1)} ... W^{(1)}; the subscripts should be ordered consistently.
  4. [§2.1 and §3.1] The PWM/PFM terminology is used inconsistently: Section 2.1 says a PFM is log-scaled to a PWM, while Section 3.1 says 'transform the PWM into a PFM [28] through the softmax operation.' The paper should clarify which matrix is converted into which.

Circularity Check

2 steps flagged · score 6.0 of 10

The transparent-model proof assumes its conclusion (p_hard_i ≤ p_free_i), and the dropout-redundancy argument defines redundancy as output invariance and then returns it as the finding.

  1. self definitional [Section 3.2, 'Constructing transparent models with prior knowledge', after Eqs. (23)-(24)]
    "Due to the limitations imposed by hard coding, it holds that p_hard_i ≤ p_free_i. ... Where ∑ m_i log(p_free_i / p_hard_i) ≥ 0, λ||W_prior−W_ref_prior||^2 ≥ 0, and γ||W_adapt||^2 − β||W_free||^2 ≈ 0. From this, it can be concluded that L_hard−L_free ≥ 0, which proves that the hard-coded model may perform worse than the unconstrained model."

    The inequality p_hard_i ≤ p_free_i is not derived from the model definitions, from optimization, or from the stated constraints; it is a pointwise version of the theorem's conclusion that hard-coded transparent models underperform. The proof then computes the loss difference and uses this same assumed inequality to conclude L_hard − L_free ≥ 0. The regularizer cancellation γ||W_adapt||^2 − β||W_free||^2 ≈ 0 is also asserted without proof. Thus the derivation is circular at its load-bearing premise.

  2. self definitional [Section 3.1, 'The redundancy of neurons', Eq. (7) and the concluding paragraph]
    "If these two neurons are redundant with respect to each other, that is, there exist parameter directions vi, vj ∈ R such that for any input x: fθ+ϵvi(x) ≈ fθ+ϵvj(x) + O(epsilon^2) ... In summary, we obtain that Δf approaches 0. That is, when redundancy occurs in neurons due to the Dropout mechanism, the output of the model does not change significantly after removing a single neuron"

    Eq. (7) does not follow from dropout training; it is an assertion that perturbations in the two removal directions give approximately equal outputs. The subsequent curvature/flatness discussion is used to conclude that removing a single neuron yields Δf ≈ 0, but that output invariance is exactly what Eq. (7) already encodes under the name 'redundant.' The proof therefore returns its own definition as the finding.

full rationale

The transparent-model theorem in Section 3.2 is circular: the pointwise inequality p_hard_i ≤ p_free_i, introduced immediately after the loss definitions, is exactly the performance-degradation claim to be proved. The subsequent algebra only repackages that assumption into L_hard − L_free ≥ 0, so the proof has no independent mathematical content. The dropout-redundancy argument is similarly definitional: Eq. (7) defines redundancy as approximate equality of outputs under perturbations in the two neuron directions, and the derivation then concludes that removing a neuron leaves the output unchanged. The rest of the theoretical section—the condition-number argument for attention and the ReLU gradient product argument—does not reduce to its inputs; those are standard, although the ReLU vanishing-gradient presentation is not fully rigorous. There is no load-bearing self-citation: reference [28] is an external review used for empirical context, not as the proof. Because the paper's advertised central contribution is rigorous mathematical derivation and at least two of the derivations assume their conclusions, the circularity score is 6.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central claims rest on multiple domain assumptions that are either unproven, such as dropout-induced redundancy and multicollinearity of keys, or effectively assume the conclusion, such as p_hard_i <= p_free_i. The free parameters epsilon and the regularization coefficients enter the proofs without specified values or derived requirements. No new physical or model entities are introduced.

free parameters (2)
  • epsilon in PWM log transform = unspecified small constant
    Introduced in Section 3.1 in the log transform of the position weight matrix; the distortion claim depends on epsilon being small, but its value and selection are not specified.
  • regularization coefficients lambda, gamma, beta = not specified
    Introduced in Eqs. 23 and 24 for the transparent-model loss comparison; the proof requires gamma times the adaptive weight norm minus beta times the free weight norm to be approximately zero, an unproved cancellation of hand-chosen coefficients.
assumptions (5)
  • domain assumption Dropout forces neurons to acquire independent features, creating redundant neurons.
    Section 3.1, 'The redundancy of neurons', states this without proof; the subsequent argument assumes redundancy and then derives output invariance.
  • ad hoc to paper Neural network output defines a smooth manifold with a connection and Riemann curvature tensor.
    Section 3.1 invokes covariant derivatives and Riemann curvature for a scalar output function f_theta(x) without defining the geometry; this is not standard and is not established.
  • domain assumption The attention key matrix is column-rank deficient due to multicollinearity.
    Section 3.2 assumes some key vector is a linear combination of others; plausible but not demonstrated for genomic sequences, and the condition number argument is applied to a matrix whose inverse attention does not compute.
  • domain assumption Fully connected layers can be abstracted into a single layer for the CNN scaling analysis.
    Section 3.1 simplifies the network structure to one fully connected layer; this ignores depth effects that the same paper analyzes in the vanishing-gradient section.
  • ad hoc to paper Hard-coded transparent models satisfy p_hard_i <= p_free_i for every sample.
    Section 3.2 asserts this as the basis for L_hard minus L_free being nonnegative; it is exactly the performance-degradation claim being proven.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Deep Learning and Explainable AI: New Pathways to Genetic Insights." pith.science (2026). https://pith.science/paper/LM63RN7D

@misc{pith2026250509873,
  author       = {Pith},
  title        = {Pith review of: Deep Learning and Explainable AI: New Pathways to Genetic Insights},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LM63RN7D}},
  note         = {Machine review of arXiv:2505.09873}
}
read the original abstract

Deep learning-based AI models have been extensively applied in genomics, achieving remarkable success across diverse applications. As these models gain prominence, there exists an urgent need for interpretability methods to establish trustworthiness in model-driven decisions. For genetic researchers, interpretable insights derived from these models hold significant value in providing novel perspectives for understanding biological processes. Current interpretability analyses in genomics predominantly rely on intuition and experience rather than rigorous theoretical foundations. In this review, we systematically categorize interpretability methods into input-based and model-based approaches, while critically evaluating their limitations through concrete biological application scenarios. Furthermore, we establish theoretical underpinnings to elucidate the origins of these constraints through formal mathematical demonstrations, aiming to assist genetic researchers in better understanding and designing models in the future. Finally, we provide feasible suggestions for future research on interpretability in the field of genetics.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 35 canonical work pages

  1. [1]

    Alipanahi, A

    B. Alipanahi, A. Delong, M. T. Weirauch, and B. J. Frey. Predicting the sequence specificities of dna-and rna-binding proteins by deep learning. Nature biotechnology, 33(8):831–838, 2015

    2015

  2. [2]

    Avsec, V

    Ž. Avsec, V . Agarwal, D. Visentin, J. R. Ledsam, A. Grabska- Barwinska, K. R. Taylor, Y . Assael, J. Jumper, P. Kohli, and D. R. Kel- ley. Effective gene expression prediction from sequence by integrating long-range interactions. Nature methods, 18(10):1196–1203, 2021

    2021

  3. [3]

    C. B. Azodi, J. Tang, and S.-H. Shiu. Opening the black box: inter- pretable machine learning for geneticists. Trends in genetics , 36(6): 442–455, 2020

    work page 2020

  4. [4]

    Bjerregaard, V

    A. Bjerregaard, V . Das, and A. Krogh. Interpretable single-cell pertur- bations from decoder gradients. In ICLR 2025 Workshop on Machine Learning for Genomics Explorations

    work page 2025

  5. [5]

    Bravo Gonzalez-Blas, I

    C. Bravo Gonzalez-Blas, I. Matetovici, H. Hillen, I. I. Taskiran, R. Van- depoel, V . Christiaens, L. Sansores-Garcia, E. Verboven, G. Hulsel- mans, S. Poovathingal, et al. Single-cell spatial multi-omics and deep learning dissect enhancer-driven gene regulatory networks in liver zona- tion. Nature Cell Biology, 26(1):153–167, 2024

    work page 2024

  6. [6]

    Dhainaut, S

    M. Dhainaut, S. A. Rose, G. Akturk, A. Wroblewska, S. R. Nielsen, E. S. Park, M. Buckup, V . Roudko, L. Pia, R. Sweeney, et al. Spatial crispr genomics identifies regulators of the tumor microenvironment. Cell, 185(7):1223–1239, 2022

    work page 2022

  7. [7]

    Dibaeinia, A

    P. Dibaeinia, A. Ojha, and S. Sinha. Interpretable ai for inference of causal molecular relationships from omics data. Science Advances, 11 (7):eadk0837, 2025

    work page 2025

  8. [8]

    Dudnyk, D

    K. Dudnyk, D. Cai, C. Shi, J. Xu, and J. Zhou. Sequence basis of tran- scription initiation in the human genome.Science, 384(6694):eadj0116, 2024

    work page 2024

Show all 44 references
  1. [9]

    H. A. Elmarakeby, J. Hwang, R. Arafeh, J. Crowdis, S. Gang, D. Liu, S. H. AlDubayan, K. Salari, S. Kregel, C. Richter, et al. Biologically informed deep neural network for prostate cancer discovery. Nature, 598(7880):348–352, 2021

    2021

  2. [10]

    X. Fu, S. Mo, A. Buendia, A. P. Laurent, A. Shao, M. d. M. Alvarez- Torres, T. Yu, J. Tan, J. Su, R. Sagatelian, et al. A foundation model of transcription across human cell types. Nature, pages 1–9, 2025

    2025

  3. [11]

    Fudenberg, D

    G. Fudenberg, D. R. Kelley, and K. S. Pollard. Predicting 3d genome folding from dna sequence with akita. Nature methods, 17(11):1111– 1117, 2020

    2020

  4. [12]

    G. E. Hinton, N. Srivastava, A. Krizhevsky, I. Sutskever, and R. R. Salakhutdinov. Improving neural networks by preventing co-adaptation of feature detectors. arXiv preprint arXiv:1207.0580, 2012

    2012 arXiv

  5. [13]

    Javed, T

    N. Javed, T. Weingarten, A. Sehanobish, A. Roberts, A. Dubey, K. Choromanski, and B. E. Bernstein. A multi-modal transformer for cell type-agnostic regulatory predictions. Cell Genomics, 2025

    2025

  6. [14]

    D. R. Kelley, J. Snoek, and J. L. Rinn. Basset: learning the regulatory code of the accessible genome with deep convolutional neural networks. Genome research, 26(7):990–999, 2016

    2016

  7. [15]

    D. R. Kelley, Y . A. Reshef, M. Bileschi, D. Belanger, C. Y . McLean, and J. Snoek. Sequential regulatory activity prediction across chromosomes with convolutional neural networks. Genome research, 28(5):739–750, 2018

    2018

  8. [16]

    S. L. Klemm, Z. Shipony, and W. J. Greenleaf. Chromatin accessibility and the regulatory epigenome. Nature Reviews Genetics , 20(4):207– 220, 2019

    2019

  9. [17]

    P. K. Koo and M. Ploenzke. Deep learning for inferring transcription factor binding sites. Current opinion in systems biology , 19:16–23, 2020

    2020

  10. [18]

    Lieberman-Aiden, N

    E. Lieberman-Aiden, N. L. Van Berkum, L. Williams, M. Imakaev, T. Ragoczy, A. Telling, I. Amit, B. R. Lajoie, P. J. Sabo, M. O. Dorschner, et al. Comprehensive mapping of long-range interactions reveals folding principles of the human genome. science, 326(5950): 289–293, 2009

    2009

  11. [19]

    Lin, Y .-C

    M.-Y . Lin, Y .-C. Lo, and J.-H. Hung. Unveiling chromatin dynamics with virtual epigenome. Nature Communications, 16(1):3491, 2025

    2025

  12. [20]

    B. Liu, Z. Jing, X. Zhang, Y . Chen, S. Mao, R. Kaundal, Y . Zou, G. Wei, Y . Zang, X. Wang, et al. Large-scale multiplexed mosaic crispr pertur- bation in the whole organism. Cell, 185(16):3008–3024, 2022

    2022

  13. [21]

    S. M. Lundberg and S.-I. Lee. A unified approach to interpreting model predictions. Advances in neural information processing systems , 30, 2017

    2017

  14. [22]

    J. Ma, M. K. Yu, S. Fong, K. Ono, E. Sage, B. Demchak, R. Sharan, and T. Ideker. Using deep learning to model the hierarchical structure and function of a cell. Nature methods, 15(4):290–298, 2018

    2018

  15. [23]

    Mathelier, O

    A. Mathelier, O. Fornes, D. J. Arenillas, C.-y. Chen, G. Denay, J. Lee, W. Shi, C. Shyr, G. Tan, R. Worsley-Hunt, et al. Jaspar 2016: a major expansion and update of the open-access database of transcription factor binding profiles. Nucleic acids research, 44(D1):D110–D115, 2016

    2016

  16. [24]

    X. Min, W. Zeng, S. Chen, N. Chen, T. Chen, and R. Jiang. Predicting enhancers with deep convolutional neural networks. BMC bioinformat- ics, 18:35–46, 2017

    2017

  17. [25]

    Molnar, G

    C. Molnar, G. Casalicchio, and B. Bischl. Interpretable machine learning–a brief history, state-of-the-art and challenges. In Joint Eu- ropean conference on machine learning and knowledge discovery in databases, pages 417–431. Springer, 2020

    2020

  18. [26]

    M. R. Mumbach, A. J. Rubin, R. A. Flynn, C. Dai, P. A. Khavari, W. J. Greenleaf, and H. Y . Chang. Hichip: efficient and sensitive analysis of protein-directed genome architecture. Nature methods, 13(11):919– 922, 2016

    2016

  19. [27]

    Naim and N

    O. Naim and N. Asher. On explaining with attention matrices. In ECAI 2024, pages 1035–1042. IOS Press, 2024

    2024

  20. [28]

    Novakovsky, N

    G. Novakovsky, N. Dexter, M. W. Libbrecht, W. W. Wasserman, and S. Mostafavi. Obtaining genetics insights from deep learning via ex- plainable artificial intelligence. Nature Reviews Genetics, 24(2):125– 137, 2023

    2023

  21. [29]

    F. F. Peleke, S. M. Zumkeller, M. Gültas, A. Schmitt, and J. Szyma´nski. Deep learning the cis-regulatory code for gene expression in selected model plants. Nature Communications, 15(1):3488, 2024

    2024

  22. [30]

    Schuette, Z

    G. Schuette, Z. Lao, and B. Zhang. Chromogen: Diffusion model pre- dicts single-cell chromatin conformations. Science Advances , 11(5): eadr8265, 2025

    2025

  23. [31]

    Shrikumar, P

    A. Shrikumar, P. Greenside, and A. Kundaje. Learning important fea- tures through propagating activation differences. In International con- ference on machine learning, pages 3145–3153. PMlR, 2017

    2017

  24. [32]

    Simonyan, A

    K. Simonyan, A. Vedaldi, and A. Zisserman. Deep inside convolutional networks: Visualising image classification models and saliency maps. arXiv preprint arXiv:1312.6034, 2013

    2013 arXiv

  25. [33]

    Sundararajan, A

    M. Sundararajan, A. Taly, and Q. Yan. Gradients of counterfactuals. arXiv preprint arXiv:1611.02639, 2016

    2016 arXiv

  26. [34]

    J. Tan, N. Shenker-Tauris, J. Rodriguez-Hernaez, E. Wang, T. Sakel- laropoulos, F. Boccalatte, P. Thandapani, J. Skok, I. Aifantis, D. Fenyö, et al. Cell-type-specific prediction of 3d chromatin organization enables high-throughput in silico genetic screening. Nature biotechno...

    2023

  27. [35]

    C. V . Theodoris, L. Xiao, A. Chopra, M. D. Chaffin, Z. R. Al Sayed, M. C. Hill, H. Mantineo, E. M. Brydon, Z. Zeng, X. S. Liu, et al. Trans- fer learning enables predictions in network biology.Nature, 618(7965): 616–624, 2023

    2023

  28. [36]

    van Hilten, S

    A. van Hilten, S. A. Kushner, M. Kayser, M. A. Ikram, H. H. Adams, C. C. Klaver, W. J. Niessen, and G. V . Roshchupkin. Gennet framework: interpretable deep learning for predicting phenotypes from genetic data. Communications biology, 4(1):1094, 2021

    2021

  29. [37]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin. Attention is all you need. Ad- vances in neural information processing systems, 30, 2017

    2017

  30. [38]

    M. D. Zeiler and R. Fergus. Visualizing and understanding convolu- tional networks. In Computer Vision–ECCV 2014: 13th European Con- ference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part I 13, pages 818–833. Springer, 2014

    2014

  31. [39]

    Zhang, C

    G. Zhang, C. Song, M. Yin, L. Liu, Y . Zhang, Y . Li, J. Zhang, M. Guo, and C. Li. Trapt: a multi-stage fused deep learning framework for pre- dicting transcriptional regulators based on large-scale epigenomic data. Nature Communications, 16(1):3611, 2025

    2025

  32. [40]

    Zhang, H

    Y . Zhang, H. Zhu, Z. Song, Y . Chen, X. Fu, Z. Meng, P. Koniusz, and I. King. Geometric view of soft decorrelation in self-supervised learn- ing. In Proceedings of the 30th ACM SIGKDD Conference on Knowl- edge Discovery and Data Mining, pages 4338–4349, 2024

    2024

  33. [41]

    Zheng, M

    A. Zheng, M. Lamkin, H. Zhao, C. Wu, H. Su, and M. Gymrek. Deep neural networks identify sequence context features predictive of tran- scription factor binding. Nature machine intelligence , 3(2):172–180, 2021

    2021

  34. [42]

    J. Zhou. Sequence-based modeling of three-dimensional genome ar- chitecture from kilobase to chromosome scale. Nature genetics, 54(5): 725–734, 2022

    2022

  35. [43]

    J. Zhou, D. R. Weinberger, and S. Han. Deep learning predicts dna methylation regulatory variants in specific brain cell types and enhances fine mapping for brain disorders. Science Advances, 11(1):eadn1870, 2025

    2025

  36. [44]

    Zuallaert, F

    J. Zuallaert, F. Godin, M. Kim, A. Soete, Y . Saeys, and W. De Neve. Splicerover: interpretable convolutional neural networks for improved splice site prediction. Bioinformatics, 34(24):4180–4188, 2018

    2018

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.