Weekly BioML Digest [July 13, 2026] (emailing issues)

Machine Learning × Computational Biology paper compilation

Hey! It's your weekly digest of machine learning papers in CompBio and Drug Discovery.

I realized that the last couple of letters got published on the web page but didn't get emailed out. I'm working on fixing the issue, stay tuned.

Feedback? Email me at biomldigest@gmail.com.

📚 Peer-Reviewed Journals (Top 20)

1347 matched filters -> 20 selected after LLM relevance + novelty ranking.

• 📡 FLOWR.ROOT – A flow matching-based foundation model for joint multi-purpose structure-aware 3D ligand generation and affinity prediction
  Cremer, Julian, Le, Tuan, ..., Menezes, Filipe, Clevert, Djork-Arné — Nature Communications, 2026-07-06
  SE(3)-equivariant flow matching unifies 3D ligand generation and affinity prediction with LoRA adaptation, enabling multiobjective, scoring-function–free design and pocket-conditional sampling.

• 🛠️ NicheTrans: spatial-aware cross-omics translation
  Wang, Zhikang, Zou, Qi, ..., Song, Jiangning, Yuan, Zhiyuan — Nature Methods, 2026-07-09
  Presents a Transformer-based, spatially aware cross-omics translation model that integrates cellular microenvironment to infer missing spatial modalities and discover disease-relevant domains (e.g., AD glial organization).

• 📰 TPS-Flow: Physics-Guided Flow-Based Generative Modeling of Protein Transition Paths.
  Kai Xu, Likun Zhao, Yanan Tian, Kewei Zhou, ..., Shihang Wang, Shaolong Lin, Huanxiang Liu, Xiaojun Yao — Journal of chemical information and modeling, 2026-07-10
  Develops a physics-guided, SE(3)-aware flow-based generative model that samples protein transition paths between states from MD-derived velocities, bridging atomistic simulation and deep generative modeling.
  Affiliations: Faculty of Applied Sciences, Macao Polytechnic University; Department of Integrative Biotechnology, Yonsei University; ...

• 📰 Cross-task interpretability through unified modeling reveals a universal shortcut bias in neoantigen prediction.
  Ziting Zhang, Lei Wei, Wenxu Wu, Hai Qi, Xiaowo Wang — Cell genomics, 2026-07-08
  Unifies peptide–MHC binding, presentation, and T-cell activation in a single transformer, revealing a universal shortcut bias from intra-HLA label imbalance and proposing a mutual-information-guided debiasing strategy.
  Affiliations: Ministry of Education Key Laboratory of Bioinformatics, Center for Synthetic and Systems Biology; Department of Basic Medical Sciences, School of Medicine; ...

• 📰 BindRNAgen: Protein-binding RNA sequence generation using latent diffusion models.
  Yan Zhou, Xiaojian Liu, Shengfan Wang, Lin Zhu, Biao Zhang, Yan Huang, Hong-Bin Shen, Xiaoyong Pan — Journal of molecular biology, 2026-07-06
  Couples a VAE with a conditional latent diffusion model to generate RNA sequences that bind input RBPs using protein language model embeddings, outperforming baselines and validated via docking/MD.
  Affiliations: Institute of Image Processing and Pattern Recognition, Shanghai Jiao Tong University; School of Health Science and Engineering, University of Shanghai for Science and Technology; ...

• 💻 A disentangled transformer-based transfer learning framework to predict patient drug response from tumor single-cell transcriptomics
  Xinliang Sun, Li Shen, Linconghua Wang, Xinyi Zhang, Zhangli Lu, Jing Tang, Min Li — Bioinformatics, 2026-07-07
  Disentangled transfer learning aligns bulk/single-cell expression to extract pharmacological signals and predict patient drug response from tumor scRNA-seq, identifying agents for drug-resistant subpopulations.

• 📰 scVAEAT: An Integrative Attention-Augmented Variational Autoencoder for Predicting Single-Cell Perturbation Responses.
  Binhua Tang, Yujia Zhang, Yiyao Chen, Xinyu Gao, Mengyao Mao — IEEE transactions on computational biology and bioinformatics, 2026-07-07
  Combines VAEs with attention-enhanced optimal transport to predict single-cell perturbation responses without paired data, accurately capturing interferon and infection signatures across cell types.

• 🖥️ Integrating cytological images and spatial transcriptomics for cell segmentation with DISSECT
  He, Yufeng, Zhao, Yanping, ..., Pan, Deng, Zeng, Zexian — Nature Computational Science, 2026-07-09
  Integrates cytology images with spatial transcriptomic gradients using a deep generative denoiser and instance-aware detection to improve single-cell segmentation and downstream spatial biology analyses.

• 📰 xBind: an integrated webserver for large language model-enabled cross-molecular protein binding site prediction.
  Xinyu Wang, Xingyue Feng, Sumit Tarafder, Debswapna Bhattacharya — Nucleic acids research, 2026-07-11
  Uses ESM-2 LLM embeddings with symmetry-aware graph neural networks to predict protein–protein/DNA/RNA binding sites from sequence or structure, with interactive, threshold-calibrated web visualization.
  Affiliations: Department of Computer Science, Virginia Tech

• 📰 ScrambleBench: a workflow for comparative assessment of structure-based de novo generative models
  Yap, Veincent, Xu, Pan, ..., Anbazhagan, Padmanabhan, Xu, Weijun — Journal of Cheminformatics, 2026-07-06
  Benchmarks structure-based generative AI across GPCRs, kinases, and hydrolases with unified metrics (e.g., HamDiv) and identifies gaps in active-site engagement and pharmacophore recognition.

• 📰 End-to-end molecular structure elucidation from multimodal NMR spectra images using vision transformers.
  Chao Han, Xiaolin Pan, Yingkai Zhang — Chemical science, 2026-07-08
  Applies a spectral Vision Transformer to raw 1D/2D NMR spectra for end-to-end structure elucidation, with simulated-to-experimental transfer improved via light fine-tuning and chemical-shift re-ranking.
  Affiliations: Department of Chemistry, New York University New York 10003 USA yingkai.zhang@nyu.edu.; Simons Center for Computational Physical Chemistry at New York University New York New York 10003 USA.; ...

• 💻 Advancing proteomic discovery through optimized multi-stage scoring and deep learning-enhanced open search
  Chen Qian, Kaifei Wang, Pengzhi Mao, Ranfei Chen, Hao Chi — Bioinformatics, 2026-07-07
  Integrates DL features into an accelerated multi-stage scoring pipeline for DDA proteomics, delivering large sensitivity gains in both restricted and open searches for PTM discovery.

• 📰 msBayesImpute as a versatile framework for addressing missing values in biomedical mass spectrometry proteomics data
  He, Jiaojiao, Helm, Barbara, ..., Klingmüller, Ursula, Lu, Junyan — Communications Chemistry, 2026-07-07
  Introduces a Bayesian factorization plus probabilistic dropout model that imputes MAR/MNAR missing proteomics values, improving normalization, differential expression, and downstream ML predictions.

• 💻 OrgNet+: towards robust protein stability prediction with convolutional neural networks
  A. Sarycheva, Aleksandr Shumilov, Petr Popov — Bioinformatics, 2026-07-07
  Trains a 3D CNN on diverse conformational ensembles to reduce prediction variance and improve ΔΔG stability estimates, yielding orientation-agnostic, ensemble-aware protein stability predictions.

• 💻 KLaR: fusing knowledge graphs and language models for biomedical target discovery
  Yinghui Jiang, Zixian Li, Yanchao Xu, Haotong Sun, Bocheng Xu, Xiangrong Liu — Bioinformatics, 2026-07-07
  Fuses local k-hop graph structure with template-textualized random-walk contexts via a gated integration of GNN and frozen sentence embeddings to enhance biomedical target/discovery link prediction.

• 💻 EPIC: Event Prototyping via Information Constrained graph learning for personalized cancer driver gene prediction
  Sang-Pil Cho, Young-Rae Cho — Bioinformatics, 2026-07-07
  Reformulates cancer driver discovery as metric learning in an event embedding space with information-constrained graph learning, prioritizing low-frequency, clinically actionable drivers.

• 💻 SIVA: diagonal integration of spatial multi-omics data via spatially informed variational autoencoders and anchor guidance
  Peng Jiang, Sishuo Chen, Xingye Wu, Juan Liu, Tian Tian — Bioinformatics, 2026-07-07
  Proposes spatially informed VAEs with Gaussian process priors and anchor guidance to diagonally integrate unpaired spatial multi-omics, outperforming existing methods across cross-slice scenarios.

• 📰 MINERVA: a public XAI-powered platform advancing multi-target discovery in Alzheimer’s disease
  Gambacorta, Nicola, Trisciuzzi, Daniela, ..., Ciriaco, Fulvio, Nicolotti, Orazio — Journal of Cheminformatics, 2026-07-06
  Builds an explainable multi-target AD platform with balanced random forests, multi-threshold bioactivity modeling, and SHAP maps for fragment-level insights across 33 AD-relevant targets.

• 📰 Reinforcement Learning-Driven Multiproperty Optimization in Molecular Design Using Multicontext Transcriptome Data.
  Yuki Matsukiyo, Chen Li, Yoshihiro Yamanishi — Journal of chemical information and modeling, 2026-07-06
  Conditions a generative model on target-gene perturbation transcriptomes within an RL loop to co-optimize drug-likeness, synthesis accessibility, and lipophilicity for bioactive molecules.
  Affiliations: Department of Complex Systems Science, Graduate School of Informatics; D3 center, The University of Osaka; ...

• 📰 PSDTA: An Approach to Drug-Target Binding Affinity Prediction by Integrating Physicochemical and Structural Information to Reduce Feature Redundancy.
  Shuang Wang, Mao Li, Peifu Han, Junteng Ma, Tianle Ma, Jingyang Ge, Na Kang, Tao Song — Journal of chemical information and modeling, 2026-07-06
  Dual-channel architecture captures both residue- and group-level binding determinants while integrating physicochemical and structural inputs, achieving SOTA DTA performance and interpretability.
  Affiliations: Qingdao Institute of Software, College of Computer Science and Technology; Faculty of Information Engineering and Automation, Kunming University of Science and Technology; ...

🧬 Preprints (arXiv + bioRxiv)

66 matched filters -> 20 selected after LLM relevance + novelty ranking.

• 🧬 FlexRibbon: Joint Sequence and Structure Pretraining for Protein Modeling
  Zhu, J.; Shi, Y.; Bi, R.; Jin, P.; ...; Wang, Y.; He, L.; Liu, H.; Qin, T. — bioRxiv, 2026-07-09
  Pretrains a joint sequence–structure protein foundation model by combining masked language modeling with diffusion-based denoising, removing MSA dependence. Achieves state-of-the-art across interface design, interaction prediction, and function tasks, especially in mutation-rich regimes.
  Affiliations: Zhongguancun Academy

• 📄 Variable-Length Generative Protein Design via Generalized Poisson Flow
  Chaoran Cheng, Zhanghan Ni, Yanru Qu, Yuxin Chen, Ruihan Guo, Jiajun Fan, Ge Liu — arXiv, 2026-07-10
  Introduces Generalized Poisson Flow for variable-length protein generation by learning an inhomogeneous event rate. Enables de novo structure/sequence design, motif scaffolding, and peptide co-design with leading conditional performance and improved designability.

• 🧬 Tokenizing single-cell transcriptomes as a native language for large language models
  Xiao, C.; Ding, Y.; Bian, H.; Chen, Y.; Wei, L.; Zhang, X. — bioRxiv, 2026-07-11
  CellTok tokenizes single-cell transcriptomes as native tokens within an LLM, unifying cellular measurements and text in a single autoregressive model. Supports cell/population analysis, disease-state inference, cell–cell communication prediction, trajectory modeling, and cellular state generation.
  Affiliations: Tsinghua University

• 🧬 Generative Drug Design in a Loop with dtSFM
  Reddy, S. T. — bioRxiv, 2026-07-08
  GenLoop closes the loop between a thermodynamics-grounded specificity foundation model and AlphaFold 3 verification to iteratively evolve small molecules. Produces high-confidence, developable designs across diverse targets using selection and counter-selection.
  Affiliations: ETH Zurich

• 🧬 CatESO: Differentiable Enzyme Sequence Optimization Guided by Substrate-Aware kcat Prediction
  Gan, Z.; Xu, Y.; Xu, J.; Wu, Z.; Huang, J.; Yin, J.; Chen, G.; Zhang, J. Z. H. — bioRxiv, 2026-07-06
  CatESO makes enzyme design differentiable by backpropagating through a substrate-aware kcat predictor under continuous sequence relaxation, regularized by ESM-2/ESMFold priors. It raises predicted turnover while maintaining foldability across stringent out-of-distribution enzymes.
  Affiliations: Shenzhen University of Advanced Technology

• 📄 MARLIN: De Novo Molecular Structure Elucidation from Tandem Mass Spectra without a Ground-Truth Formula
  Xujun Che, Xiuxia Du, Depeng Xu — arXiv, 2026-07-06
  MARLIN performs de novo molecular structure elucidation directly from MS/MS without a ground-truth formula using a self-supervised fingerprint encoder and a block-diffusion language model with a safe mass-shell constraint. Achieves state-of-the-art exact-match and formula recovery on NPLIB1.

• 🧬 Computational design of de novo integrated domains enables rational control of pathogen effector recognition in plant NLR immune receptors.
  Xi, Y.; Bucknell, A. H.; Watson, J. L.; Maqbool, A.; ...; Emmrich, P. M. F.; Talbot, N. J.; Banfield, M. J.; Bentham, A. R. — bioRxiv, 2026-07-10
  Uses RFdiffusion and ProteinMPNN to design de novo integrated domains that, when inserted into plant NLRs, confer programmable effector recognition and immune signaling. Demonstrates cross-pathogen targeting with structural/biophysical validation, enabling rational crop immunity engineering.
  Affiliations: Centre for Programmable Biological Matter, Department of Biosciences

• 🧬 TMO: ASYMMETRIC CROSS-MODAL ATTENTION FOR LEARNING CELL-STATE-DEPENDENT REGULATORY LAGS FROM SINGLE-CELL MULTIOMIC DATA
  Lopez-Delgado, P. A.; Delgado-Carlo, M. M. — bioRxiv, 2026-07-08
  TMO employs asymmetric cross-modal attention with a lag prior to learn signed, cell-state-dependent ATAC→RNA regulatory lags from single-cell multiome data. Delivers near-perfect lag concordance and is validated by TF ChIP-seq and Perturb-seq across tissues.
  Affiliations: Universidad Nacional Autónoma de México

• 📄 Predicting Therapeutic Outcome via Aligning Patient-Specific Knowledge Graph and Gene-Level Perturbation Representations
  Dongmin Bang, Sugyun An, Inyoung Sung, Ilho Yun, Sun Kim, Sangseon Lee — arXiv, 2026-07-06
  PREDIKTOR aligns patient-specific gene-regulatory knowledge-graph embeddings with simulated drug perturbation transcriptomes from a LINCS-pretrained model via contrastive learning. Improves drug-response prediction under patient/drug/tissue splits and transfers zero-shot to I-SPY2.

• 📄 Accurate, Interdisciplinary and Transparent Structure-property Understanding with Deep Native Structural Reasoning
  Chen Tang, Yizhou Wang, Jianyu Wu, Lintao Wang, ..., Phil Torr, Bowen Zhou, Wanli Ouyang, Lei Bai — arXiv, 2026-07-08
  SciReasoner discretizes 3D structures into a shared vocabulary to perform native structural reasoning across proteins, small molecules, and crystals. Provides interpretable evidence chains and broad SOTA on 67/86 tasks, including GO annotation and retrosynthesis.

• 📄 Canopy: A Heterograph Foundation Model for Metabolic Engineering
  Jake Bowden, Laurence Legon, Satnam Surae — arXiv, 2026-07-07
  Canopy is a heterograph foundation model unifying sequences, chemistry, text, and experiments in a multi-modal knowledge graph with HGT pretraining. Frozen embeddings substantially improve fermentation titer prediction for metabolic engineering over tabular and homogeneous GNN baselines.

• 🧬 Designing Fidelity of CRISPR-Cas Endonucleases by Kinetic Insights
  Liu, H.; Zhou, Z.; Yuan, L.; Pang, B.; ...; Ren, R.; Warshel, A.; Lei, Y.; Zhu, L. — bioRxiv, 2026-07-07
  Combines PLM-guided residue prioritization with atomistic MD and automated pathway search to resolve pre-cleavage kinetic transitions in CRISPR nucleases. Pinpoints fidelity-diminishing residues and yields ultra-high-specificity Cas9/Cas12a variants with minimal wet-lab screening.
  Affiliations: The Chinese University of Hong Kong, Shenzhen

• 📄 Autoregressive latent diffusion for 3D molecule generation
  Federico Ottomano, Gaopeng Ren, Yingzhen Li, Kim E. Jelfs, Alex M. Ganose — arXiv, 2026-07-10
  KRONOS is a latent autoregressive diffusion model that jointly generates molecular topology and 3D geometry while supporting fragment-conditioned fill-in. Achieves leading autoregressive performance and competes with diffusion models on QM9 and GEOM-Drugs.

• 📄 Rethinking Benchmarks and Models for Enzyme Specificity Prediction
  Elizabeth H. Mahood, Natália Komorníková, Tomáš Pluskal, Pranam Chatterjee — arXiv, 2026-07-06
  Rebenchmarks enzyme–substrate/reaction specificity models in discovery-relevant regimes and shows many fail to generalize beyond training enzymes. Adapts Boltz complex embeddings to ES prediction, surpassing BLAST on a large CYP ranking benchmark.

• 📄 TheBioCollection: Unified Pre-Training Scale LLM Corpus for Biology
  Hyunjin Seo, Hyeon Hwang, Gyubok Lee, Jay Shin, ..., Sanghoon Lee, Hongjoon Ahn, Sungjun Han, Sangwon Jung — arXiv, 2026-07-09
  TheBioCollection assembles a 52.6B-token, multi-entity biological corpus with derived properties and a matched evaluation suite. Pretraining on it more than doubles a 16B LLM’s score on biology tasks across molecules, proteins, genomes, cells, and pathways with minimal loss of general language ability.

• 🧬 Benchmarking AlphaFold and related deep learning approaches for modeling antibody and TCR antigen recognition
  Yin, R.; Saravanakumar, S.; Shi, S. Y.; Park, M.; ...; Felbinger, N.; Kaufman, S.; Eisenberg, M.; Pierce, B. — bioRxiv, 2026-07-06
  Systematically benchmarks AlphaFold2/3 and sampling strategies on antibody–protein/peptide and TCR–pMHC complexes. Finds AF3 and increased sampling improve success variably by interface class and that pooling models lifts antibody–peptide near-native hits.
  Affiliations: University of Maryland

• 📄 A Quiet Failure in Calibrated Virtual Screening: Marginal Conformal Prediction Under-Covers the Minority Class, and a Class-Conditional Fix Recovers It
  Muhammadjon Tursunbadalov, Mustafojon Tursunbadalov — arXiv, 2026-07-07
  Shows marginal conformal prediction under-covers the minority (active) class in imbalanced virtual screening despite correct overall coverage. Class-conditional (Mondrian) conformal restores per-class guarantees with modest increases in prediction-set size.

• 🧬 PEPstrMOD2: Next-generation tertiary structure prediction of chemically modified and non-natural peptides
  Jain, S.; Mehta, N. K.; Raina, S.; Kumar, P.; Varun,; Raghava, G. P. S. — bioRxiv, 2026-07-06
  PEPstrMOD2 integrates AF2/ESMFold initialization with expanded AMBER-compatible libraries to model peptides with PTMs, non-canonicals, D-forms, and cyclic constraints. Achieves competitive accuracy to AF3 on modified/cyclic benchmarks at scale.
  Affiliations: Indraprastha Institute of Information Technology Delhi

• 🧬 Variational Autoencoder-enabled High-throughput Drug Screening for HIV Latency Modulators predicted through Noise in Gene Expression
  Shukla, D.; Lu, Y.; Horne, J. R.; Mi, X.; Nag, S.; Dash, S.; Dar, R. D. — bioRxiv, 2026-07-09
  Trains a variational autoencoder on time-lapse single-cell fluorescence to infer noise features predictive of HIV latency modulation. An in silico screen of ~175k compounds yields validated LRA synergizers and LPAs with a 15.9% experimental hit rate.
  Affiliations: University of Illinois at Urbana-Champaign

• 🧬 A control-validated pan-proteome deep-learning pipeline nominates GPR35 as a candidate target of the orphan bacterial metabolite ligiamycin A
  Martin, J. — bioRxiv, 2026-07-06
  A pan-proteome cross-attentional DTI model with bias-corrected ranking nominates class-A GPCRs as targets of the orphan natural product ligiamycin A. Control-anchored docking prioritizes GPR35 as an experimentally testable mechanism.
  Affiliations: Rowan-Virtua School of Osteopathic Medicine