Pith. sign in

REVIEW 5 major objections 5 minor 61 references

Modeling and prediction of mutation fitness on protein functionality with structural information using high-dimensional Potts model

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A node-wise sparse-group Potts estimator gives provable convergence rates for protein mutation fitness prediction.

desk verdict The paper has a genuinely new estimator and a real empirical win, but the headline convergence rate doesn't actually cover the K=20 setting where it's applied. read the letter →

arxiv 2505.14958 v1 pith:F6IPORXU submitted 2025-05-20 stat.ME

Bingying Dai , Yinan Lin , Kejue Jia , Zhao Ren , Wen Zhou This is my paper · ORCID
classification stat.ME MSC 62J0762F1262H12
keywords PottsmodelsparsegroupLassonode-wisemultinomialregressionproteinmutationfitnessevolutionaryenergystructuralinformationmultiplesequencealignmenthigh-dimensionalstatistics
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 claims that protein mutation effects can be predicted from aligned sequence data by fitting a high-dimensional Potts model through node-wise multinomial regression with a sparse group Lasso penalty. The penalty simultaneously selects which residue sites interact and which amino acid pairs carry the interaction, and structural distances between sites enter as group weights. The paper establishes non-asymptotic $\ell_1$ and $\ell_2$ convergence rates for the estimated parameters, matching the minimax lower bound for double-sparse linear models up to a factor that depends only on the multinomial, 20-amino-acid nature of the model. Because plug-in energy changes inherit these rates, the model's mutation-fitness predictions come with a consistency guarantee. On 12 protein families, the method's predicted energy changes show higher Spearman correlation with experimental mutation fitness than the EVM benchmark in ten of the twelve families.

What carries the argument

The machinery is node-wise multinomial regression: for each site $j$, the amino acid state at $j$ is the $K$-category response and the states at all other sites are covariates, so the Potts conditional probability becomes a multinomial logistic model. The objective adds a sparse group Lasso penalty that combines a group penalty across site-pair blocks (selecting interacting sites) with an element-wise $\ell_1$ penalty within each block (selecting amino-acid-level couplings). Structural information enters as group weights $w_{jr}$ built from $\alpha$-carbon distances. The theoretical engine is a refined bound on the stochastic term $\sum_{k=1}^K \epsilon_k^{\top} X u_k$, adapted from sharp oracle inequalities for high-dimensional linear models, which avoids incoherence-type conditions and yields the $\ell_1$ and $\ell_2$ error bounds in Theorem 3.1 and the consistency of estimated energy changes in Corollary 3.2.

What would settle it

Simulate MSAs from a known Potts model where rows are sampled along a phylogenetic tree rather than independently, then fit the estimator and compare the empirical $\ell_2$ error to the rate in (13) as $n$ grows; if the error fails to shrink at the claimed $n^{-1/2}$ rate, the independence assumption in Assumption 1 is load-bearing for the convergence claim.

Watch

Extended reading notes

Core claim

The central claim is that a node-wise sparse-group-Lasso multinomial regression provides a computationally tractable estimator of Potts-model parameters whose $\ell_2$ error converges at the same rate as the minimax lower bound for high-dimensional linear models with a sparse group structure, up to a factor $R_K$ determined by the multinomial log-likelihood (conditional variance, minimum conditional probability, and minimum covariance eigenvalue). This is, to the authors' knowledge, the first such convergence guarantee for the high-dimensional Potts model, and it extends to plug-in estimates of single- and multiple-site evolutionary energy changes. Empirically, the paper reports that incorporating structural distances as group weights improves prediction of experimentally measured mutation fitness and outperforms EVM in ten of twelve protein families.

Load-bearing premise

The convergence theorem assumes the aligned sequences are independent draws with a well-conditioned covariance and no near-zero conditional amino acid probabilities; real multiple sequence alignments are phylogenetically correlated, and the paper's sequence reweighting and rare-category exclusions fall outside the theorem's coverage.

Editorial extensions

If this is right

  • The $\ell_2$ error bound in (13) implies that, with probability tending to one, the squared-error loss of the estimated dependence parameters is $O_P\big((s^\circ_g \log(d/s^\circ_g) + s^\circ \log(K^2 s^\circ_g/s^\circ))/n\big)$, so energy-change predictions are consistent as sequence number grows.
  • Because the bound is tighter than an $\ell_1$-only or group-only penalty when the truth is doubly sparse, the estimator's advantage carries over to any Potts-type categorical graphical model with such sparsity.
  • Structural group weights improve prediction: in Table 1, the method beats EVM in ten of twelve families and beats unweighted sparse-group Lasso in most cases, indicating spatial proximity is informative for mutation fitness.
  • The node-wise formulation parallelizes across sites and avoids sampling the partition function, making the method scalable to MSAs with hundreds of sites and thousands of sequences.

Reading between the lines

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

  • A natural extension would be to learn the distance-to-coupling transform $K(D)$ from an independent structure dataset and then apply it to a new family, which could make the structural-weighting claim predictive rather than fitted per family.
  • Since the theorem's factor $R_K$ degrades with $K$, grouping the 20 amino acids into a smaller alphabet of biochemical classes is a testable way to improve finite-sample accuracy, at the cost of atomistic resolution.
  • The same double-sparse node-wise scheme could be applied to other categorical graphical models, such as gene regulatory networks from allele or expression data, wherever group and element sparsity are both plausible.
  • The robustness of the weighting scheme to two different $K(\cdot)$ forms in simulations suggests that exact distance normalization is not critical, but this was only tested under distance-dependent generative models.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes a node-wise high-dimensional multinomial regression with a sparse group Lasso penalty to estimate Potts-model parameters from protein multiple sequence alignments, with protein structural distances incorporated through group weights. It claims non-asymptotic l1 and l2 convergence rates for the estimator, consistency results for estimated mutation energy changes, and Spearman correlations of predicted energy changes with experimental mutation fitness for 12 protein families, arguing that the method outperforms EVM and other baselines. The paper also includes simulation studies, two protein case studies, and a limitations section.

Significance. If valid, the l2 rate result would be a useful contribution to high-dimensional multinomial regression and to Potts-model estimation, and the proposed estimator addresses a practically relevant problem. The paper is also to be credited for providing reproducible code, for testing the robustness of the spatial kernel choice in simulations, and for candidly listing limitations. However, the claimed theoretical guarantee does not cover the K=20 settings used in the simulations and real-data analyses, and the spatial-kernel form is chosen using four families that later appear in the validation table. As a result, the current contribution is best viewed as a practically motivated method plus an asymptotic theory for regimes with very large n relative to K, rather than a theory that directly explains the reported protein-fitness results.

major comments (5)
  1. [Section 3, Theorem 3.1] The theorem's sparsity condition is not satisfied by any K=20 experiment reported. Because the K+1 conditional probabilities in Assumption 2 sum to one, c_* is at most 1/(K+1). Since the authors assume all theta_jk are nonzero, s^o is at least K. For K=20, the condition s^o << c_*^2 sqrt(n/log(dK)) requires n >> s^o^2/c_*^4 * log(dK) >= 20^2 * 21^4 * log(dK) ~ 7.8e7 * log(dK). The largest real-data sample in Table 1 has n ~ 1.2e5 and the simulations in Table 2 have n <= 4000, so the l2 bound (12) is not in force for the estimators whose energy changes are reported. The abstract's statement that the theorem enables accurate quantification of estimated energy changes is therefore unsupported in the demonstrated regime.
  2. [Section 4, Eqs. (14)-(15); Table 1] The spatial kernel K(D_jr)=1-exp(-D^2/MS_j) is determined in Section 4 by inspecting the dependency-versus-distance relationship for DYR, TRY2, FYN, and YAP, and the same four families appear in the validation Table 1 (rows DYR, TRY2, FYN, YAP1). For these families the claimed improvement over EVM is partly an in-sample comparison. The robustness check with an alternative kernel (N2) in Section 6.2 is useful but is conducted on simulated data, not on the real families; the real-data claim across 12 protein families should be qualified, or the kernel should be fixed before inspecting any validation family.
  3. [Section 3.2 and Limitations] The paper states that the l2 rate 'matches' the minimax lower bound for high-dimensional linear models with sparse group structure, but the Limitations section concedes that no minimax lower bound is established for multinomial regression over the double-sparse class. Matching a linear-model lower bound is not an optimality statement for the Potts model, and the phrase 'matching the existing minimax lower bound' in the abstract should be revised to avoid implying Potts-model optimality.
  4. [Supplementary Material references] The proof of Theorem 3.1 is said to rely on Lemma B1 and other results in the Supplementary Material, and the numerical sections refer to Section C.1 and Table C1 of the Supplementary Material, but no supplement is included with this preprint. The central proof therefore cannot be checked, and the claimed rates are not verifiable in the current version; a complete supplement should accompany the manuscript.
  5. [Table 1] Table 1 reports Spearman correlations without confidence intervals or any measure of uncertainty. Several differences are small (BLAT 0.65 vs 0.57, DLG4 0.55 vs 0.54, MTH3 0.52 vs 0.51), so the statement that the method outperforms EVM in ten of twelve families is not statistically grounded. Bootstrap or split-sample intervals should be provided for the correlations.
minor comments (5)
  1. [Section 2.1, Eq. (3)] The identifiability paragraph says 'we assign k=0 to the wild-type amino acid', but Section 2.1 earlier defines k=0 as the alignment gap. If the wild-type category is used as the baseline, the gauge transformation that absorbs the gamma_jr,k,a_r terms into theta_jk should be stated explicitly; as written, the relationship between Eq. (3) and the fitted parameters is ambiguous.
  2. [Section 5, sequence reweighting] The sample weights omega_i in Section 5 are computed from the same MSA, while Theorem 3.1 assumes equally weighted iid rows. The sentence that the results 'readily extend' is an assertion, not a proof; the mismatch should be acknowledged or the extension stated as a formal corollary.
  3. [Algorithm 1] In the gradient update line, the notation [nabla^2 ell(gamma_hat_j) nabla ell(gamma_hat_j)]_(r) appears to be a typo; it should probably be [nabla^2 ell(gamma_hat_j)(gamma_j - gamma_hat_j)]_(r) or similar.
  4. [Table 2] Table 2 reports Monte Carlo averages but no standard errors; adding them would help assess whether differences between methods are meaningful.
  5. [Section 5.2] The text uses 'E.coli'; the conventional form is 'E. coli'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the theoretical bounds are conditional derivations, and the empirical claims rest on external fitness benchmarks rather than on the fitted inputs.

full rationale

The paper's theoretical contribution (Theorem 3.1 and Corollaries 3.2–3.3) is a conditional non-asymptotic analysis. The ℓ1/ℓ2 bounds are derived from Assumptions 1–2 using oracle inequalities from Bellec et al. (2018) and Li et al. (2023); no fitted parameter is later renamed as a prediction. The energy-change consistency bound in Corollary 3.2 follows by direct triangle inequality from the ℓ1 error of the parameter estimates, not by definitional equivalence. The empirical comparison in Table 1 is anchored to externally measured mutagenesis fitness (Tm, SEL, abundance, etc.), so the main empirical claim is not self-referential. The kernel K(D_jr) = 1 − exp(−D^2_jr/MS_j) was selected after inspecting DYR, TRY2, FYN, and YAP, and those four families also appear in Table 1; this is a data-dependent model-selection choice, but it is not a circular reduction because the validation target (experimental fitness) was not used to select K, and the method still outperforms EVM in the remaining eight independent families. The only self-citation (Jernigan et al., 2021) supports the biological premise of site-wise sparsity; it is not load-bearing for the rate theorem. The Limitations section concedes that no minimax lower bound for multinomial regression itself has been established and that sample/group weights may be derived from the same dataset; these are scope and correctness caveats, not circularity. The concern that Theorem 3.1's sparsity condition may fail for the K = 20 experiments is a potential mismatch between theorem coverage and numerical settings, not evidence that the derivation reduces to its own inputs. Overall, no circular step can be exhibited from the paper's own equations or citations.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The estimator is a penalized nodewise regression, so its validity rests on tuning constants, sparsity and design assumptions, and the externally chosen spatial kernel. The theory itself uses no invented entities; the main uncharged inputs are the i.i.d. design assumption and the bounded-probability assumption.

free parameters (4)
  • Lasso tuning lambda = selected by 5-fold CV in practice; theory sets lambda=2*lambda_sharp
    The estimator (9) and the rate (12) depend on lambda; theory requires lambda proportional to lambda_sharp, which depends on unknown sparsity and variance. No data-independent choice is provided.
  • Group Lasso tuning lambda_g = sqrt(s/s_g)*lambda
    Same as above; lambda_g depends on unknown sparsity levels s and s_g.
  • Hamming distance reweighting threshold = 0.2
    Sample weights in Section 5 use threshold 0.2 on normalized Hamming distance, following Morcos et al.; this is a hand-chosen constant affecting real data estimates.
  • Spatial kernel form K(D_jr) = 1 - exp(-D^2/MS_j)
    The functional form of K relative to distance was selected by inspecting fitted group Lasso norms versus distance in four protein families (Section 4). It is not derived from theory and is then used in the evaluation.
assumptions (5)
  • domain assumption Rows {x_{-j}^{(i)}} are iid with covariance Sigma satisfying lambda_min(Sigma) >= c_lambda (Assumption 1).
    Invoked in Theorem 3.1; MSA sequences with phylogenetic structure likely violate iid.
  • domain assumption Bounded logits and minimum conditional probability c* > 0 (Assumption 2).
    Needed for quadratic expansion of multinomial loss; rare amino acids and gaps in MSA make c* near zero.
  • domain assumption True dependency vectors are (s,s_g)-sparse with s, s_g fixed and sparsity conditions in Theorem 3.1.
    Central to rate; if true interactions are dense, sparse group penalty results are not covered.
  • ad hoc to paper Spatially close sites have stronger direct couplings, expressed through K(D_jr)=1-exp(-D^2/MS_j).
    Used to construct group weights (14); form selected based on four protein families, not derived from first principles.
  • domain assumption Theoretical results extend to estimated sample weights and structural weights with only a factor 1/w_min change (Remark 4.1, Section 5).
    Stated without proof; weights in real data are estimated from the same MSA and structure rather than independent.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Modeling and prediction of mutation fitness on protein functionality with structural information using high-dimensional Potts model." pith.science (2026). https://pith.science/paper/F6IPORXU

@misc{pith2026250514958,
  author       = {Pith},
  title        = {Pith review of: Modeling and prediction of mutation fitness on protein functionality with structural information using high-dimensional Potts model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/F6IPORXU}},
  note         = {Machine review of arXiv:2505.14958}
}
abstract

Quantifying the effects of amino acid mutations in proteins presents a significant challenge due to the vast combinations of residue sites and amino acid types, making experimental approaches costly and time-consuming. The Potts model has been used to address this challenge, with parameters capturing evolutionary dependency between residue sites within a protein family. However, existing methods often use the mean-field approximation to reduce computational demands, which lacks provable guarantees and overlooks critical structural information for assessing mutation effects. We propose a new framework for analyzing protein sequences using the Potts model with node-wise high-dimensional multinomial regression. Our method identifies key residue interactions and important amino acids, quantifying mutation effects through evolutionary energy derived from model parameters. It encourages sparsity in both site-wise and amino acid-wise dependencies through element-wise and group sparsity. We have established, for the first time to our knowledge, the $\ell_2$ convergence rate for estimated parameters in the high-dimensional Potts model using sparse group Lasso, matching the existing minimax lower bound for high-dimensional linear models with a sparse group structure, up to a factor depending only on the multinomial nature of the Potts model. This theoretical guarantee enables accurate quantification of estimated energy changes. Additionally, we incorporate structural data into our model by applying penalty weights across site pairs. Our method outperforms others in predicting mutation fitness, as demonstrated by comparisons with high-throughput mutagenesis experiments across 12 protein families.

Figures

Figures reproduced from arXiv: 2505.14958 by the authors.

Figure 1
Figure 1. (a) MSA data for the DYR family at sites 143-153. (b) Protein structure highlighting the spatial proximity of sites 143 and 150, with [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The distances 𝐷𝑗𝑟 between sites 𝑗 and 𝑟 versus ∥b𝜸 (𝑔) 𝑗(𝑟) ∥2 with the fitted trend. Remark 4.1. When group weights are considered, the theoretical results in Section 3 can be easily extended to cases where the weights are fixed or derived from an independent dataset, such as the protein’s structural data, by adjusting 𝜆♯ of (11) to 𝜆♯/𝑤min, where 𝑤min = min𝑟≠𝑗,𝑤𝑗𝑟≠0 𝑤𝑗𝑟. As 𝑤min is independent of the sample size o… view at source ↗
Figure 3
Figure 3. Predicted mutation fitness of DYR. (a) Landscape of estimated energy changes for different amino acids occurring at various sites. (b) [PITH_FULL_IMAGE:figures/full_fig_p022_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Predicted mutation fitness of DLG4. (a) Landscape of estimated energy change for different amino acids at each sites. The framed [PITH_FULL_IMAGE:figures/full_fig_p024_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

61 extracted references · 61 canonical work pages

  1. [1]

    Abramovich, F., Grinshtein, V., and Levy, T. (2021). Multiclass classification by sparse multinomial logistic regression. IEEE Trans. Inform. Theory , 67(7):4637--4646

    work page 2021

  2. [2]

    A., Schmidt, S., Peshkin, L., Ramensky, V

    Adzhubei, I. A., Schmidt, S., Peshkin, L., Ramensky, V. E., Gerasimova, A., Bork, P., Kondrashov, A. S., and Sunyaev, S. R. (2010). A method and server for predicting damaging missense mutations. Nat. Methods. , 7(4):248--249

    work page 2010

  3. [3]

    L., Fowler, D

    Araya, C. L., Fowler, D. M., Chen, W., Muniez, I., Kelly, J. W., and Fields, S. (2012). A fundamental protein property, thermodynamic stability, revealed solely from large-scale measurements of protein function. Proc. Natl. Acad. Sci. , 109(42):16858--16863

    work page 2012

  4. [4]

    Baccanari, D., Stone, D., and Kuyper, L. (1981). Effect of a single amino acid substitution on escherichia coli dihydrofolate reductase catalysis and ligand binding. J. Biol. Chem. , 256(4):1738--1747

    work page 1981

  5. [5]

    G., Lee, S.-I., and Langmead, C

    Balakrishnan, S., Kamisetty, H., Carbonell, J. G., Lee, S.-I., and Langmead, C. J. (2011). Learning generative models for protein fold families. Proteins. , 79(4):1061--1078

    work page 2011

  6. [6]

    C., Lecu\'e, G., and Tsybakov, A

    Bellec, P. C., Lecu\'e, G., and Tsybakov, A. B. (2018). Slope meets L asso: improved oracle bounds and optimality. Ann. Statist. , 46(6B):3603--3642

    work page 2018

  7. [7]

    Bershtein, S., Mu, W., and Shakhnovich, E. I. (2012). Soluble oligomerization provides a beneficial fitness effect on destabilizing mutations. Proc. Natl. Acad. Sci. , 109(13):4857--4862

    work page 2012

  8. [8]

    and Riordan, O

    Bollob \'a s, B. and Riordan, O. (2011). Sparse graphs: metrics and random models. Random Struct. Algorithms , 39(1):1--38

    work page 2011

Show all 61 references
  1. [9]

    T., Li, H., Ma, J., and Xia, Y

    Cai, T. T., Li, H., Ma, J., and Xia, Y. (2019). Differential markov random field analysis with an application to detecting differential microbial community networks. Biometrika , 106(2):401--416

    2019

  2. [10]

    T., Zhang, A

    Cai, T. T., Zhang, A. R., and Zhou, Y. (2022). Sparse group lasso: optimal sample complexity, convergence rate, and statistical inference. IEEE Trans. Inform. Theory , 68(9):5975--6002

    2022

  3. [11]

    Consortium, T. U. (2020). Uniprot: the universal protein knowledgebase in 2021. Nucleic. Acids. Res. , 49(D1):D480--D489

    2020

  4. [12]

    Friedman, J., Hastie, T., and Tibshirani, R. (2010). Regularization paths for generalized linear models via coordinate descent. J. Stat. Softw. , 33(1):1–22

    2010

  5. [13]

    Gao, J., Pang, X., Zhang, L., Li, S., Qin, Z., Xie, X., and Liu, J. (2023). Transcriptome analysis reveals the neuroprotective effect of dlg4 against fastigial nucleus stimulation-induced ischemia/reperfusion injury in rats. BMC Neurosci. , 24(1):40

    2023

  6. [14]

    S., and Chen, J

    Guo, Z., Rakshit, P., Herman, D. S., and Chen, J. (2021). Inference for the case probability in high-dimensional logistic regression. J. Mach. Learn. Res. , 22:Paper No. [254], 54

    2021

  7. [15]

    A., Ingraham, J

    Hopf, T. A., Ingraham, J. B., Poelwijk, F. J., Sch \"a rfe, C. P., Springer, M., Sander, C., and Marks, D. S. (2017). Mutation effects predicted from sequence co-variation. Nat. Biotechnol. , 35(2):128--135

    2017

  8. [16]

    Huang, J., Ma, S., and Zhang, C.-H. (2008). Adaptive lasso for sparse high-dimensional regression models. Statist. Sinica , 18(4):1603--1618

    2008

  9. [17]

    Ising, E. (1925). Beitrag zur theorie des ferromagnetismus. Z. Physik , 31(1):253--258

    1925

  10. [18]

    Izenman, A. J. (2021). Sampling algorithms for discrete M arkov random fields and related graphical models. J. Amer. Statist. Assoc. , 116(536):2065--2086

    2021

  11. [19]

    Jernigan, R., Jia, K., Ren, Z., and Zhou, W. (2021). Large-scale multiple inference of collective dependence with applications to protein function. Ann. Appl. Stat. , 15(2):902--924

    2021

  12. [20]

    T., Buchan, D

    Jones, D. T., Buchan, D. W., Cozzetto, D., and Pontil, M. (2012). Psicov: precise structural contact prediction using sparse inverse covariance estimation on large multiple sequence alignments. Bioinformatics , 28(2):184--190

    2012

  13. [21]

    M., Jain, P., O'roak, B

    Kircher, M., Witten, D. M., Jain, P., O'roak, B. J., Cooper, G. M., and Shendure, J. (2014). A general framework for estimating the relative pathogenicity of human genetic variants. Nat. Genet. , 46(3):310--315

    2014

  14. [22]

    O., Starita, L

    Kitzman, J. O., Starita, L. M., Lo, R. S., Fields, S., and Shendure, J. (2015). Massively parallel single-amino-acid mutagenesis. Nat. Methods. , 12(3):203--206

    2015

  15. [23]

    J., Minikel, E

    Lek, M., Karczewski, K. J., Minikel, E. V., Samocha, K. E., Banks, E., Fennell, T., O’Donnell-Luria, A. H., Ware, J. S., Hill, A. J., Cummings, B. B., et al. (2016). Analysis of protein-coding genetic variation in 60,706 humans. Nature , 536(7616):285--291

    2016

  16. [24]

    M., Haldane, A., and Flynn, W

    Levy, R. M., Haldane, A., and Flynn, W. F. (2017). Potts hamiltonian models of protein co-variation, free energy landscapes, and evolutionary fitness. Curr. Opin. Struct. Biol. , 43:55--62

    2017

  17. [25]

    and Abramovich, F

    Levy, T. and Abramovich, F. (2023). Generalization error bounds for multiclass sparse linear classifiers. J. Mach. Learn. Res. , 24:Paper No. [151], 35

    2023

  18. [26]

    and Luan, Y

    Li, H. and Luan, Y. (2003). Kernel cox regression models for linking gene expression profiles to censored survival data. Pac. Symp. Biocomput. , pages 65--76

    2003

  19. [27]

    Li, Q., Yi, F., Wang, T., Xiao, G., and Liang, F. (2017). Lung cancer pathological image analysis using a hidden potts model. Cancer Inform. , 16:1176935117711910

    2017

  20. [28]

    Li, Z., Zhang, Y., and Yin, J. (2023). Sharp minimax optimality of LASSO and SLOPE under double sparsity assumption. Available at https://arxiv.org/abs/2308.09548

    2023 arXiv

  21. [29]

    Lounici, K., Pontil, M., van de Geer, S., and Tsybakov, A. B. (2011). Oracle inequalities and optimal inference under group sparsity. Ann. Statist. , 39(4):2164--2204

    2011

  22. [30]

    T., and Li, H

    Ma, R., Guo, Z., Cai, T. T., and Li, H. (2022). Statistical inference for genetic relatedness based on high-dimensional logistic regression. Available at https://arxiv.org/abs/2202.10007

    2022 arXiv

  23. [31]

    S., Colwell, L

    Marks, D. S., Colwell, L. J., Sheridan, R., Hopf, T. A., Pagnani, A., Zecchina, R., and Sander, C. (2011). Protein 3d structure computed from evolutionary sequence variation. PLOS ONE , 6(12):e28766

    2011

  24. [32]

    S., Hopf, T

    Marks, D. S., Hopf, T. A., and Sander, C. (2012). Protein structure prediction from sequence variation. Nat. Biotechnol. , 30(11):1072--1080

    2012

  25. [33]

    N., Poelwijk, F

    McLaughlin Jr., R. N., Poelwijk, F. J., Raman, A., Gosal, W. S., and Ranganathan, R. (2012). The spatial architecture of protein function and adaptation. Nature , 491(7422):138--142

    2012

  26. [34]

    L., Gamble, C

    Melamed, D., Young, D. L., Gamble, C. E., Miller, C. R., and Fields, S. (2013). Deep mutational scanning of an rrm domain of the saccharomyces cerevisiae poly(a)-binding protein. RNA , 19(11):1537--1551

    2013

  27. [35]

    Melnikov, A., Rogov, P., Wang, L., Gnirke, A., and Mikkelsen, T. S. (2014). Comprehensive mutational scanning of a kinase in vivo reveals substrate-dependent fitness landscapes. Nucleic. Acids. Res. , 42(14):e112

    2014

  28. [36]

    N., Reeves, R., and Berthelsen, K

    M ller, J., Pettitt, A. N., Reeves, R., and Berthelsen, K. K. (2006). An efficient markov chain monte carlo method for distributions with intractable normalising constants. Biometrika , 93(2):451--458

    2006

  29. [37]

    S., Sander, C., Zecchina, R., Onuchic, J

    Morcos, F., Pagnani, A., Lunt, B., Bertolino, A., Marks, D. S., Sander, C., Zecchina, R., Onuchic, J. N., Hwa, T., and Weigt, M. (2011). Direct-coupling analysis of residue coevolution captures native contacts across many protein families. Proc. Natl. Acad. Sci. , 108(49):E1293--E1301

    2011

  30. [38]

    Ng, P. C. and Henikoff, S. (2003). Sift: predicting amino acid changes that affect protein function. Nucleic. Acids. Res. , 31(13):3812--3814

    2003

  31. [39]

    and Hastie, T

    Nibbering, D. and Hastie, T. J. (2022). Multiclass-penalized logistic regression. Comput. Statist. Data Anal. , 169:Paper No. 107414, 16

    2022

  32. [40]

    and Haran, M

    Park, J. and Haran, M. (2018). Bayesian inference in the presence of intractable normalizing functions. J. Amer. Statist. Assoc. , 113(523):1372--1390

    2018

  33. [41]

    Potts, R. B. (1952). Some generalized order-disorder transformations. Proc. Cambridge Philos. Soc. , 48:106--109

    1952

  34. [42]

    P., Gerrow, K., Wang, Y

    Prange, O., Wong, T. P., Gerrow, K., Wang, Y. T., and El-Husseini, A. (2004). A balance between excitatory and inhibitory synapses is controlled by psd-95 and neuroligin. Proc. Natl. Acad. Sci. , 101(38):13915--13920

    2004

  35. [43]

    and Li, T

    Qi, M. and Li, T. (2024). The non-overlapping statistical approximation to overlapping group lasso. J. Mach. Learn. Res. , 25(115):1--70

    2024

  36. [44]

    J., and Lafferty, J

    Ravikumar, P., Wainwright, M. J., and Lafferty, J. D. (2010). High-dimensional ising model selection using _1 -regularized logistic regression. Ann. Statist. , 38(3):1287--1319

    2010

  37. [45]

    and Amini, A

    Razaee, Z. and Amini, A. (2020). The potts-ising model for discrete multivariate data. In Adv. Neural Inf. Process. Syst. , volume 33, pages 13727--13737. Curran Associates, Inc

    2020

  38. [46]

    Rockah-Shmuel, L., T \'o th-Petr \'o czy, \'A ., and Tawfik, D. S. (2015). Systematic mapping of protein mutational space by prolonged drift reveals the deleterious effects of seemingly neutral mutations. PLoS. Comput. Biol. , 11(8):e1004421

    2015

  39. [47]

    R., Dyson, H

    Schnell, J. R., Dyson, H. J., and Wright, P. E. (2004). Structure, dynamics, and catalytic function of dihydrofolate reductase. Annu. Rev. Biophys. Biomol. Struct. , 33:119--140

    2004

  40. [48]

    I., Dicker, A

    Schweitzer, B. I., Dicker, A. P., and Bertino, J. R. (1990). Dihydrofolate reductase as a therapeutic target. FASEB J. , 4(8):2441--2452

    1990

  41. [49]

    Simon, N., Friedman, J., Hastie, T., and Tibshirani, R. (2013). A sparse-group lasso. J. Comput. Graph. Statist. , 22(2):231--245

    2013

  42. [50]

    Song, D., Wang, Q., Yan, G., Liu, T., Sun, T., and Li, J. J. (2024). sc D esign3 generates realistic in silico data for multimodal single-cell and spatial omics. Nat. Biotechnol. , 42(2):247--252

    2024

  43. [51]

    M., Pruneda, J

    Starita, L. M., Pruneda, J. N., Lo, R. S., Fowler, D. M., Kim, H. J., Hiatt, J. B., Shendure, J., Brzovic, P. S., Fields, S., and Klevit, R. E. (2013). Activity-enhancing mutations in an e3 ubiquitin ligase identified by high-throughput mutagenesis. Proc. Natl. Acad. Sci. , 11...

    2013

  44. [52]

    and Feng, Y

    Tian, Y. and Feng, Y. (2023). Transfer learning under high-dimensional generalized linear models. J. Amer. Statist. Assoc. , 118(544):2684--2697

    2023

  45. [53]

    V., and Feng, Y

    Tian, Y., Rusinek, H., Masurkar, A. V., and Feng, Y. (2024). _1 -penalized multinomial regression: estimation, inference, and prediction, with an application to risk factor identification for different dementia subtypes. Available at https://arxiv.org/abs/2302.02310

    2024 arXiv

  46. [54]

    Tibshirani, R. (1996). Regression shrinkage and selection via the lasso. J. Roy. Statist. Soc. Ser. B , 58(1):267--288

    1996

  47. [55]

    and Gertheiss, J

    Tutz, G. and Gertheiss, J. (2016). Regularized regression for categorical data. Stat. Model. , 16(3):161--200

    2016

  48. [56]

    van de Geer, S., B\"uhlmann, P., Ritov, Y., and Dezeure, R. (2014). On asymptotically optimal confidence regions and tests for high-dimensional models. Ann. Statist. , 42(3):1166--1202

    2014

  49. [57]

    and Hansen, N

    Vincent, M. and Hansen, N. R. (2014). Sparse group lasso and high dimensional multinomial classification. Comput. Statist. Data Anal. , 71:771--786

    2014

  50. [58]

    Wainwright, M. J. (2019). High-Dimensional Statistics: A Non-Asymptotic Viewpoint , volume 48 of Cambridge Series in Statistical and Probabilistic Mathematics . Cambridge University Press, Cambridge, first edition

    2019

  51. [59]

    N., and Lu, J

    Wang, J., Lu, H., Plataniotis, K. N., and Lu, J. (2009). Gaussian kernel optimization for pattern classification. Pattern Recognit. , 42(7):1237--1247

    2009

  52. [60]

    and Lin, Y

    Yuan, M. and Lin, Y. (2006). Model selection and estimation in regression with grouped variables. J. R. Stat. Soc. Ser. B Stat. Methodol. , 68(1):49--67

    2006

  53. [61]

    and Li, Y

    Zhang, J. and Li, Y. (2023). High-dimensional gaussian graphical regression models with covariates. J. Amer. Statist. Assoc. , 118(543):2088--2100

    2023

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.