Weekly BioML Digest [June 29, 2026]

Machine Learning × Computational Biology paper compilation

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

Feedback? Email me at biomldigest@gmail.com.

📚 Peer-Reviewed Journals (Top 20)

1100 matched filters -> 20 selected after LLM relevance + novelty ranking.

• 🧫 Efficient generation of epitope-targeted antibodies with Germinal
  Mille-Fragoso, Luis S., Driscoll, Claudia L., ..., Hie, Brian L., Gao, Xiaojing J. — Nature Biotechnology, 2026-06-23
  Germinal co-optimizes antibody sequence and structure with a structure predictor plus an antibody-specific protein language model, enabling de novo CDR design against specified epitopes with high nanomolar binding success across multiple antigens. Demonstrates low-n experimental validation and broad applicability to therapeutic antibody discovery.

• 🏛️ Linear-time prediction of proteome-scale microbial protein interactions.
  Andre Cornman, Matt Tranzillo, Nicolo G Zulaybar, Imane Bouzit, Yunha Hwang — Proceedings of the National Academy of Sciences of the United States of America, 2026-06-23
  FlashPPI is a contrastive-learning framework that embeds microbial proteins in a shared latent space for linear-time, proteome-scale prediction of physical PPIs from sequence alone. It matches or outperforms structure-based pipelines while cutting screening from days to minutes, enabling rapid interactome mapping for microbial biology.
  Affiliations: Tatta Bio, Cambridge; Microbiology Graduate Program, Massachusetts Institute of Technology; ...

• 📰 Tumor-naïve ctDNA detection with deep learning-enhanced error suppression for sensitive mutation calling
  Akbarinejad, Shaya, Doppler, Sarah, ..., Ibn-Salem, Jonas, Weber, David — Genome Medicine, 2026-06-23
  DEEPctMUT integrates UMIs with machine learning and deep learning–based error suppression to sensitively call tumor-naïve ctDNA mutations down to 0.03% VAF. It outperforms commercial tumor-naïve assays and improves performance panel-independently, enabling earlier, minimally invasive cancer detection and monitoring.

• 📰 High-throughput machine learning-aided antibody discovery for cell surface antigens.
  Deepash Kothiwal, Aaron W Kollasch, Murali Anuganti, Nicholas Hollmer, ..., Shaotong Zhu, Timothy A Springer, Debora S Marks, Rob Meijers — Cell systems, 2026-06-23
  A synthetic Fab yeast display library tailored for ML seamlessly links selection data to models, yielding hundreds of cell-surface antigen binders and enabling ML-driven binder identification for new targets (e.g., ROBO2, PD-L2). Provides an open, ML-compatible dataset and workflow to accelerate antibody discovery and development.
  Affiliations: Institute for Protein Innovation, Boston; Department of Systems Biology, Harvard Medical School; ...

• 📰 DeepPath: overcoming data scarcity for protein transition pathway prediction using physics-based deep learning.
  Yui Tik Pang, Lixinhao Yang, Katie M Kuo, James C Gumbart — Chemical science, 2026-06-23
  DeepPath combines physics-guided deep learning with generative active learning to rapidly predict atomistic protein transition pathways and refine them using force-field oracles. It reproduces biologically relevant intermediates (e.g., BAM hybrid-barrel state) and transient interactions, advancing dynamic structural biology.
  Affiliations: School of Physics, Georgia Institute of Technology Atlanta GA 30332 USA gumbart@physics.gatech.edu.; School of Chemistry and Biochemistry, Georgia Institute of Technology Atlanta GA 30332 USA.

• 🚀 Long-stranded XNA-cssDNA hybrids for robust data storage
  Xinyu Sun, Yufeng Pei, P. Tan, Tianyuan Bian, ..., Yan Wang, Jiachen Xie, Liang Hong, Jie Song — Science Advances, 2026-06-24
  A temperature-guided language model was used to evolve a FANA polymerase (Tgomut) ~4.4× faster than TgoD4K, enabling synthesis of >7.5 kb XNA strands for robust DNA/XNA hybrid data storage. Demonstrates LLM-driven protein engineering for synthetic biology and biotechnology.

• 📰 pyVIPER: a fast and scalable Python package for protein activity estimation and master regulator analysis of single-cell RNA sequencing data
  Wang, Alexander L. E., Zanella, Luca, ..., Califano, Andrea, Vasciaveo, Alessandro — BMC Bioinformatics, 2026-06-22
  pyVIPER brings GPU-accelerated, Python-based protein activity inference to scRNA-seq with modular encoders and scverse compatibility, enabling rapid, network-based single-cell profiling. Scales VIPER analyses to large atlases.

• 📰 Transformer-accelerated discovery of inhibitors targeting the RpsA_Δ438 deletion in PZA-resistant tuberculosis
  Xiao, Jiazhuo, Shahab, Muhammad, Zhang, Haoke, Huang, Zunnan — Journal of Cheminformatics, 2026-06-22
  A Transformer-based virtual screening pipeline, coupled with docking, GaMD, and MM/GBSA, identifies inhibitors targeting the RpsA_Δ438 mutant in PZA-resistant Mycobacterium tuberculosis. The top hit shows stable binding and favorable ADMET, illustrating LLM-enabled, resistance-focused TB drug discovery.

• 📰 Machine learning reveals proteome-encoded growth determinants underlying metabolic versatility of Rhodopseudomonas palustris on lignin-derived aromatics.
  Abraham Osinuga, Mark Kathol, Rajib Saha — mSystems, 2026-06-23
  CorePredX integrates proteomics with explainable ML to uncover a compact, nonredundant set of proteins predicting growth across oxygen/substrate conditions in Rhodopseudomonas palustris. It reveals a hierarchical growth architecture linking translation, redox buffering, and storage cycling, guiding metabolic engineering for lignin valorization.
  Affiliations: Department of Chemical and Biomolecular Engineering, University of Nebraska-Lincoln

• 📰 Integrated machine learning and molecular dynamics–driven multi-target virtual screening of FDA-approved drugs for drug repurposing in breast cancer
  Kırboğa, Kevser Kübra, Acar, Şükran, Şen, Metehan, Rudrapal, Mithun — In Silico Pharmacology, 2026-06-22
  A multimodal framework combines machine learning classification, ensemble docking, and multi-target MD/MM-GBSA to repurpose FDA drugs against HER2, EGFR, VEGFR2, HDAC3, and CDK6 in breast cancer. Ponatinib and entrectinib emerged as polypharmacology leads with stable binding across targets.

• 📰 Transcending Structural Dependencies: A Tunable Mass Spectrometry-Driven Machine Learning Framework for Genotoxicity Prediction.
  Wanhao Sun, Min Li, Zhijian Wang, Huangjiange Cai, ..., Fei Xu, Xiangxiang Zeng, Hongru Feng, Yuanjiang Pan — Environmental science & technology, 2026-06-23
  GenoToxMass is a mass spectrometry–driven, structure-independent ML pipeline for genotoxicity screening that achieves AUC 0.95 on curated spectra. It shows >85% concordance with Ames test outcomes in real samples and provides SOP-ready, tunable thresholds for regulatory deployment.
  Affiliations: Department of Chemistry, Zhejiang University; Institute of Fundamental and Transdisciplinary Research, Zhejiang University; ...

• 📰 Computer-Aided Rational Modification of Hst 5 Based on Ssa1/2 and the Antifungal Activity of the Derivatives against Candida spp.
  Yiqing Sun, Ruifang Li, Kedong Yin, Jinhua Zhang, ..., Hong-Jie Yang, Shuangshuang Bi, Shiyu Li, Shengxian Chen — Journal of medicinal chemistry, 2026-06-23
  A CADD/AI-guided workflow (ESM2-AFPpred screening, docking, alanine scanning) rationally redesigned Histatin-5 into Hst5-22-RW with improved antifungal potency and uptake via Ssa1/2 in Candida albicans. Demonstrates AI-enabled optimization of antimicrobial peptides.

• 📰 A Hybrid Experimental and in silico Platform for ITPK1 Chemical Probe Discovery.
  Adam Yasgar, Sankalp Jain, Huanchen Wang, Chih-Shia Lee, ..., Robin E Stanley, Alexey V Zakharov, Ji Luo, Natalia J Martinez — SLAS discovery : advancing life sciences R & D, 2026-06-23
  A hybrid experimental–in silico platform couples 1,536-well HTS with ML and pharmacophore modeling to identify selective ITPK1 inhibitors, including crystallographically validated hits. Provides a scalable, ML-augmented pipeline for probe discovery in inositol phosphate signaling.
  Affiliations: National Center for Advancing Translational Sciences, National Institutes of Health; National Center for Advancing Translational Sciences, National Institutes of Health; ...

• 📰 Sequence-structure cross-attention model integrating ESM-2 embeddings and AlphaFold cues for accurate prediction of Escherichia coli protein solubility.
  Z Elmi — Computers in biology and medicine, 2026-06-23
  SeqStruct-XAttn fuses ESM-2 embeddings with AlphaFold-derived structural cues via cross-attention to predict E. coli protein solubility (R² up to 0.575 ensemble; AUC 0.920 binary). The sequence–structure integration yields calibrated, transferable predictions across taxa.
  Affiliations: Department of Software Engineering, Beykoz University

• 📰 Breast Cancer Diagnosis and HER2+ Versus Triple Negative Discrimination by Infrared Spectral Histopathology.
  Hayat El Tahech, Cédric Lerévérend, Seydou Kane, Caroline Fichel, ..., Jean-Hugues Salmon, Olivier Piot, Stéphane Potteaux, Cyril Gobinet — Analytical chemistry, 2026-06-23
  An FTIR + ML platform classifies breast cancer and discriminates HER2+ vs TNBC from tissue spectra with high accuracy; SHAP maps spectral features to biochemical differences. Offers rapid, label-free histopathology.
  Affiliations: Université de Reims Champagne-Ardenne, BioSpecT UR7506; Université de Reims Champagne-Ardenne, IRMAIC UR7509; ...

• 📰 MTA-Swin: A Multi-Token Attention Swin Transformer for Brain Tumor Classification with Leakage-Free MRI Benchmarking
  Lu, Dong, Zhang, Yu, Chaudhary, Divya — Journal of Medical Systems, 2026-06-22
  MTA-Swin introduces multi-token attention into Swin Transformers and, after leakage-free curation of a large MRI dataset, achieves 98.6% accuracy in brain tumor subtype classification. Demonstrates robust, interpretable vision architecture for clinical neuro-oncology.

• 📰 A highly interpretable machine learning model for predicting lung cancer bone metastasis: uncovering the synergistic effect of routine biochemical markers
  Jiang, Zi-Feng, Ke, Zhang-Yan, ..., Fu, Jin-Bao, Zhang, Yan-Bei — Clinical and Translational Oncology, 2026-06-22
  An interpretable GBDT model using routine biochemistry predicts bone metastasis in lung cancer (AUC 0.81) and highlights a synergistic non-linear effect of ALP with D-dimer. Provides transparent, low-cost risk stratification to inform imaging and management.

• 📰 Voxel-Wise Radiomics Habitat Analysis of Posttreatment Gliomas for Noninvasive Differentiation of True Progression and Pseudoprogression.
  Linsha Yang, Defeng Liu, Duo Zhang, Juan Du, Qinglei Shi, Tao Zheng — Journal of magnetic resonance imaging : JMRI, 2026-06-23
  Voxel-wise habitat radiomics from multiparametric MRI, fused with clinical and molecular data, improves differentiation of true progression vs pseudoprogression in gliomas (AUC 0.89 external). SHAP identifies perfusion–edge heterogeneity as key, enabling interpretable decision support.

• 📰 Translational diffusion and isomerization reaction of a liquid crystal molecule at solid-liquid interface of ionic liquids studied by total internal reflection-transient grating spectroscopy.
  Masaki Fujiwara, Hideaki Shirota, Kento Kasahara, Nobuyuki Matubayasi, Yoshifumi Kimura — Soft matter, 2026-06-23
  A deep learning framework restores Raman spectra transmitted through scattering biological tissue by learning an inverse transformation from paired pre/post-transmission data. Achieves >95% cosine similarity to ground truth, enhancing quantitative molecular diagnostics in vivo.
  Affiliations: Graduate School of Science and Engineering, Doshisha University; Department of Chemistry, Chiba University; ...

• 📰 Deep Learning-Based Restoration of Distorted Transmission Raman Spectra through Biological Tissue.
  Haoqiang Xie, Zhou Chen, Yutong Zhou, Zehou Su, Zongyu Wu, Linley Li Lin, Jian Ye — Analytical chemistry, 2026-06-23
  A 1D U-Net model trained on paired pre/post-transmission Raman/SERS spectra across tissues restores intensity and spectral profiles degraded by scattering, improving concentration quantification. Enables accurate, through-tissue molecular readouts for biomedical Raman spectroscopy.
  Affiliations: School of Biomedical Engineering, Shanghai Jiao Tong University; Institute of Medical Robotics, Shanghai Jiao Tong University; ...

🧬 Preprints (arXiv + bioRxiv)

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

• 🧬 HTS-Oracle X: AI-Guided Prospective Discovery of Small Molecule Immune Checkpoint Binders
  Abdel-Rahman, S.; Gabr, M. — bioRxiv, 2026-06-22
  Introduces a multimodal deep learner (cross-attention ChemBERTa + RDKit, trained on continuous binding signals with MC-dropout UQ) that prospectively discovers small-molecule binders to immune checkpoints (CD28, TIM-3, VISTA), achieving a 30% hit rate with multiple sub-µM actives.
  Affiliations: Weill Cornell Medicine

• 🧬 A complete RXFP1-relaxin interaction model unlocks the design of potent mini-protein modulators
  Clement, J.; Lkhagvajargal, T.; Hoare, B. L.; Myint, T.; ...; Wang, C.; Knott, G. J.; Bathgate, R. A.; Grinter, R. — bioRxiv, 2026-06-22
  Combines deep learning structural modeling of the RXFP1–relaxin complex with de novo mini-protein design to yield the first potent, selective agonists and antagonists (low-nanomolar) for this GPCR, elucidating an activation mechanism and enabling structure-guided modulation.
  Affiliations: Univeristy of Melbourne

• 📄 Sesame: Structure-Aware Molecular Generation via Spatial Density-Map Conditioning
  Konstantin Yatsenko, Arvind Thiagarajan — arXiv, 2026-06-22
  Presents Sesame, a diffusion model with a spatial pairformer that conditions jointly on protein-pocket and partial-molecule density maps, enabling de novo ligand generation and fragment-conditioned lead optimization within a single framework.

• 📄 Scalable Peptide Design via Memory-Efficient Equivariant Transformer
  Rui Jiao, Xiangzhe Kong, Yinjun Jia, Yijia Zhang, Ziyi Yang, Yang Liu, Jianzhu Ma — arXiv, 2026-06-23
  Introduces MEET, a memory-efficient E(3)-equivariant transformer backbone for full-atom peptide sequence–structure co-design, integrated into VAE/diffusion pipelines to scale with atom count and improve binding affinity, physical validity, and diversity.

• 📄 PerturbCellRL: Verifier-Guided Reinforcement Learning for Single-Cell Perturbation Prediction
  Dongxia Wu, Mingyu Li, Yuhui Zhang, Anurendra Kumar, Emma Lundberg, Serena Yeung-Levy, Emily B. Fox — arXiv, 2026-06-26
  Proposes PerturbCellRL, which post-trains a single-cell generative model via verifier-guided reinforcement learning using cell-level rewards (similarity, DE, pathway activity), yielding biologically consistent, pathway-aligned perturbation predictions.

• 🧬 GenoME: a MoE-based generative model for individualized, multimodal prediction and perturbation of genomic profiles
  Wei, J.; Xue, Y.; Chai, H.; Gao, Y. Q. — bioRxiv, 2026-06-23
  GenoME is a Mixture-of-Experts generative model that uses DNA sequence and single-cell ATAC-seq to predict unified epigenomic, transcriptomic, and 3D chromatin profiles, generalizing to unseen cell types and accurately forecasting multimodal effects of in silico perturbations.
  Affiliations: Changping Laboratory

• 🧬 CoLa-VAE: A Cell-Cell Communication-Aware Variational Autoencoder for Representation Learning and Expression Denoising
  Chen, Y.; Qi, C.; Fang, H.; Luan, F.; Zhang, Z.; Arya, S.; Wei, Z. — bioRxiv, 2026-06-26
  CoLa-VAE incorporates ligand–receptor-derived cell–cell communication graphs via dynamic Laplacian regularization to jointly learn denoised scRNA-seq and latent representations, enhancing communication program detection, batch mitigation, and spatial deconvolution.
  Affiliations: New Jersey Institute of Technology

• 📄 Autoregressive Boltzmann Generators
  Danyal Rehman, Charlie B. Tan, Yoshua Bengio, Avishek Joey Bose, Alexander Tong — arXiv, 2026-06-25
  Autoregressive Boltzmann Generators replace normalizing-flow BGs with autoregressive models for equilibrium sampling, improving accuracy and scalability on peptide systems and introducing Robin, a transferable 132M-parameter sampler.

• 📄 Uncertainty-aware reinforcement learning for chemical language models
  Borja Medina, Jon Paul Janet — arXiv, 2026-06-23
  Adds uncertainty awareness to RL-driven chemical language models by optimizing for and modulating updates with predictive uncertainty, yielding more reliable exploration and higher true hit rates without sacrificing objective scores.

• 🧬 Rett syndrome lifespan extension in mice via AI-guided ADAR editing
  Savva, Y. A.; Booth, B. J.; Shumaker, L.; Fasnacht, R.; ...; Works, M. G.; Huss, D. J.; Briggs, A. W.; VanSchoiack, A. A. — bioRxiv, 2026-06-28
  Uses generative AI–designed ADAR-recruiting gRNAs to correct MECP2 R168X at the RNA level, restoring MeCP2 in patient neurons and rescuing phenotypes with ~70% editing and extended lifespan in a Rett syndrome mouse model.
  Affiliations: Shape Therapeutics

• 📄 Deciphering Fingerprints of 3D Molecular Surfaces for Accurate Epitope Prediction
  Fang Wu, Weihao Xuan, Jure Leskovec, Yejin Choi, Li Erran Li — arXiv, 2026-06-22
  SurfBind learns directly on 3D molecular surfaces with binder-aware cross-attention and hierarchical coarse-to-fine prediction, achieving state-of-the-art antibody epitope prediction and generalization across unseen antibodies and conformations.

• 📄 3D Masked Autoencoders are Robust Learners of Volumetric and Multimodal Cellular Representations for Microscopy
  Amirhossein Kardoost, Lion Gleiter, Tingying Peng, Carsten Marr — arXiv, 2026-06-22
  Demonstrates that 3D masked autoencoders on volumetric fluorescence microscopy, aligned with protein language models, yield superior single-cell representations and state-of-the-art protein localization and PPI prediction.

• 🧬 Fault-tolerant 3D reconstruction from 2D spatial proteomics sections
  Zhang, Z.; Tan, Y.; Nolan, G.; Snyder, M.; Ma, Z. — bioRxiv, 2026-06-28
  3D-Omics-Flow is a generative pipeline that repairs and interpolates sparse 2D spatial proteomics sections to reconstruct single-cell–resolution 3D tissue volumes, enabling robust atlas construction from imperfect stacks.
  Affiliations: Yale University

• 📄 Neural operator-based digital twins for modeling amyloid-$β$ and tau propagation and treatment optimization in Alzheimer's disease
  Xiaofeng Xu, Tingting Dan, Zifan Zhou, Bin Li, Guorong Wu, Wenrui Hao — arXiv, 2026-06-23
  Learns patient-specific reaction–diffusion dynamics of amyloid-β and tau via neural operators from longitudinal PET, then optimizes PDE-constrained therapies to personalize intervention strategies in Alzheimer’s disease.

• 🧬 BATTLE-AMP: Benchmarking Antimicrobial Peptide Predictors
  Szymczak, P.; Bukała, A.; Zarzecki, W.; Sala, M.; ...; Gambin, A.; L. Müller, C.; Setny, P.; Szczurek, E. — bioRxiv, 2026-06-24
  BATTLE-AMP benchmarks AMP predictors against MIC data across species, showing MIC-trained models outperform binary classifiers and exposing unresolved activity cliffs and composition-indistinguishable failures.
  Affiliations: Institute of AI for Health, Helmholtz Zentrum München; Faculty of Mathematics, Informatics and Mechanics

• 🧬 Multi-Scale Machine Learning for Antibody-Antigen Binding Affinity Prediction Using Deep Mutational Scanning and Structural Features
  Sivasubramani, S. — bioRxiv, 2026-06-23
  Fuses physicochemical, structural, PLM, and SASA/ΔΔGfold features to predict mutation impacts on Ab–Ag affinity under leave-one-complex-out DMS, with a confidence-stratified ensemble attaining 79% of the theoretical performance ceiling.
  Affiliations: IIT DELHI

• 🧬 Scaling SMILES-Based Chemical Language Models for Therapeutic Peptide Engineering
  Feller, A. L.; Secor, M.; Swanson, S.; Wilke, C. O.; Deibler, K. — bioRxiv, 2026-06-23
  PeptideCLM-2 scales SMILES-based chemical language models to natively represent complex peptide chemistry, outperforming prior approaches on key developability endpoints like membrane diffusion, function, and half-life.
  Affiliations: The University of Texas at Austin; Novo Nordisk

• 📄 ConSolv: Solvent-Conditional Machine Learning Implicit Solvent Potential
  Linying Zhang, Julija Zavadlav — arXiv, 2026-06-23
  ConSolv introduces a solvent-conditional ML potential with attention-based solvent embeddings, accurately predicting solvation free energies across 66 organic solvents and generalizing to unseen environments and NMR observables.

• 🧬 Fluorescence Blinking Patterns Fingerprint the Local Protein Environment
  Püntener, S.; Kossmann, D.; Bielec, K.; Rivera-Fuentes, P. — bioRxiv, 2026-06-23
  Leverages single-molecule fluorescence blinking kinetics of a covalent label and deep learning to fingerprint local protein environments, enabling identification of proteins, binding pockets, and single PTMs with mechanistic interpretability.
  Affiliations: University of Zurich

• 🧬 kontakteUR: transforming coordinates to chemical intuition to focus on essential interactions in biomolecular systems
  Scherlo, M.; Wippermann, E.; Fuertges, T.; Kuenne, R.; ...; Ruetten, F.; Boeckmann, M.; Hoeweler, U.; Rudack, T. — bioRxiv, 2026-06-22
  Transforms 3D structures into a standardized contact-space representation capturing chemically informed residue–residue interactions over time, enabling scalable comparison and AI-ready features for complexes and design.
  Affiliations: Structural Bioinformatics, University of Regensburg