Weekly BioML Digest [July 28, 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)

1197 matched filters -> 20 selected after LLM relevance + novelty ranking.

• 📰 Abstract A030: Patient-level prediction of trial outcomes with a calibrated pan-cancer foundation model
  D. Bertolini, F. Fuller, J. Christopher, Jonathan R. Walsh, S. Liang, Aaron M. Smith — Clinical Cancer Research, 2026-07-21
  Builds a calibrated pan-cancer transformer foundation model that predicts patient-level survival and comparative effectiveness by aligning RWD-derived generative cohorts to trial landmarks, reproducing Phase III outcomes and enabling synthetic control arms.

• 📰 DeorphaNN: Virtual screening of GPCR peptide agonists using AlphaFold-predicted active-state complexes and deep learning embeddings.
  Larissa Ferguson, Sébastien Ouellet, Elke Vandewyer, Christopher Wang, Zaw Wunna, Tony K Y Lim, William R Schafer, Isabel Beets — Molecular cell, 2026-07-23
  Introduces DeorphaNN, a graph neural network that integrates AlphaFold active-state GPCR–peptide complexes and deep embeddings to prioritize endogenous peptide agonists, experimentally deorphanizing receptors across species.
  Affiliations: Neurobiology Division, MRC Laboratory of Molecular Biology; Independent Researcher, Ottawa; ...

• 📡 DirectContacts2: a wiring diagram of human physical protein interactions
  Claussen, Erin R., Woodcock-Girard, Miles D., Fischer, Samantha N., Drew, Kevin — Nature Communications, 2026-07-24
  Trains a machine-learning classifier on >25,000 proteomics experiments to distinguish direct from indirect human protein interactions, enabling high-confidence structural modeling and disease-complex wiring maps.

• 📰 Quantum convolutional HLA immunogenic peptide prediction (Q-CHIPP): Next-generation neoantigen prediction with quantum neural network.
  Ryan Peters, Kahn Rhrissorrakrai, Prerana Bangalore Parthasarathy, Vadim Ratner, ..., Laxmi Parida, Sara Capponi, Filippo Utro, Tyler J Alban — Science advances, 2026-07-24
  Demonstrates quantum convolutional neural networks for MHC binding and immunogenicity with noise-mitigation on real hardware, introducing Q-CHIPP to improve neoantigen prioritization with fewer training samples than classical models.
  Affiliations: Center for Immunotherapy and Precision Immuno-Oncology, Cleveland Clinic; Cleveland Clinic Research, Cleveland Clinic; ...

• 📰 Application of vision transformers to protein-ligand affinity prediction
  Poziemski, Jakub, Siedlecki, Pawel — Scientific Reports, 2026-07-25
  Applies vision transformers to 3D protein–ligand affinity prediction, capturing global long-range interactions and outperforming state-of-the-art baselines while revealing biologically relevant attention patterns.

• 📰 Decoding cryptic defluorinases through a latent generative sequence landscape.
  Ke Ji, Sydney S Barnes, Cheyenne Ziegler, Marjan Nikpey, ..., Elizabeth K Pack, Nikita Kvasovs, Faruck Morcos, Sheel C Dodani — Chemical science, 2026-07-22
  Uses a latent generative landscape trained on enzyme families to decode cryptic defluorinases; experimentally validates highly thermostable defluorinases with superior catalytic profiles over known enzymes.
  Affiliations: Department of Chemistry and Biochemistry, The University of Texas at Dallas Richardson TX 75080 USA sheel.dodani@utdallas.edu.; Department of Biological Sciences, The University of Texas at Dallas Richardson TX 75080 USA faruckm@utdallas.edu.; ...

• 📰 AlphaDTA: integrating AlphaFold3 embeddings and 3D complex structures for drug–target binding affinity prediction
  Chung, Minjae, Park, Sejin, Lee, Hyunju — Journal of Cheminformatics, 2026-07-21
  AlphaDTA integrates AlphaFold3 complex structures and single/pair embeddings with a 3D geometric encoder to predict drug–target affinity under low-overlap splits, enabling target-specific repurposing case studies.

• 💻 RelAgent: A multi-agent solution for molecular relationship grounding.
  Rubing Chen, Jiaxin Wu, C. Zhang, Xiaoyong Wei, Qing Li — Bioinformatics, 2026-07-23
  RelAgent is a multi-agent framework that grounds natural-language molecular relations to precise substructures using extraction, localization, and ontology-guided reasoning, sharply boosting performance on MolGround.

• 📰 Deep Learning Prediction of O‐Glycopeptide Tandem Mass Spectra Enhances O‐Glycoproteomics
  Yu Zong, Yuxin Wang, Liang Qiao — Advanced Science, 2026-07-23
  DeepGPO combines Transformers with graph neural networks and pretraining to predict O‑glycopeptide MS/MS spectra, enabling site localization even without site-determining ions and supporting mono/double O‑glycoforms.

• 📰 Rapid and energy-efficient ultra-large library screening for drug discovery on a SpiNNaker2 neuromorphic chip
  Jimenez Siegert, Johnny Alexander, Kelber, Florian, ..., Mayr, Christian, Meiler, Jens — Communications Chemistry, 2026-07-21
  Implements ligand-based ultra-large virtual screening on a SpiNNaker2 neuromorphic chip, achieving 60% higher throughput and 86% lower energy than an embedded GPU while scanning a 19-billion compound space.

• 📰 Abstract B050: Novel, conformational target discovery in TKI-resistant non-small cell lung cancer
  Patric W. Sadecki, A. Ritter, Hetal D Marble, Min Hak Lee, Byoung Chul Cho, N. Goodwin, Daniel Benjamín, Faraz Choudhary — Clinical Cancer Research, 2026-07-21
  Identifies conformationally altered surfaceome targets in TKI‑resistant NSCLC using deep structural proteomics and GNN-driven reasoning, nominating ADC-ready SPC targets beyond overexpression.

• 📰 Abstract A031: Large language models ensemble deciphers spatial proteogenomic landscapes to identify a novel trop2-cd47 co-targeting axis in non-small cell lung cancer
  Aakash Desai, S. Alhushki, E. McNeeley, J. Deshane, Kenneth P. Hough, Kayla F. Goliwas — Clinical Cancer Research, 2026-07-21
  An LLM ensemble integrates spatial transcriptomics and proteomics to uncover a TROP2–CD47 co-targeting axis in NSCLC, with cross-model agreement gating to mitigate bias and hallucination.

• 📰 Abstract A076: Seeing the Drug and Seeing What It Is Doing: Integrated Spatial PK/PD Analysis for Hybrid and Next-Generation Oncology Therapeutics
  Andrew Yatsuhashi, Gargey Yagnik, Phillip Carvalho, Ziying Liu, ..., Letao Ma, Xinli Liu, Kenneth J. Rothschild, Mark J Lim — Clinical Cancer Research, 2026-07-21
  Presents a multiomic spatial PK/PD workflow that co-registers drug localization with proteins/RNA in the same tissue, enabling mechanism-aware efficacy readouts for next-gen hybrid therapeutics.

• 💻 PathMED: An R toolkit for single-sample molecular scoring and machine learning with omics data.
  Jordi Martorell-Marugán, Ivan Ellson, R. López-Domínguez, Pablo Pedro Jurado-Bascón, ..., Chang Wang, Frédéric Baribaud, D. Toro-Domínguez, P. Carmona-Sáez — Bioinformatics, 2026-07-24
  Releases pathMED, an R/Bioconductor toolkit that unifies pathway scoring with ML to train cross-omics prognostic models, demonstrating transcriptome-to-proteome deployment and pathway dissection.

• 📰 Large language models unlock large text corpora in the search for data on medicinal plants and fungi
  Adam Richard‐Bollans, Francesco Civita, T. Cossu, Ifra Saifi, K. Patmore, Caroline Wilkinson, B. Allkin — PLANTS, PEOPLE, PLANET, 2026-07-23
  Shows that LLMs (GPT‑4o et al.) can extract structured ethnopharmacology relations from massive text corpora with high precision, extending medicinal plant/fungi datasets for Kew’s Plants for Health.

• 📰 Clinical AI-assisted FXR prioritization and WADDAICA-guided design of a berberine-inspired analogue for MASH: integrated computational and experimental validation
  Hui Yao, Yifeng Zhou, Huijie Zhang, Man Ni, Yang Liu, Yuan Zhou — Frontiers in Pharmacology, 2026-07-23
  Combines clinical feature-learning (Transformer) with WADDAICA-guided molecular design to prioritize FXR in MASH and propose a berberine-inspired analogue validated by MD/DFT and in vitro assays.

• 📡 cellGeometry: ultra-fast single-cell deconvolution of bulk RNA-Seq using a geometric solution
  Lau, Rachel, Çubuk, Cankut, ..., Pitzalis, Costantino, Lewis, Myles J. — Nature Communications, 2026-07-23
  Introduces cellGeometry, a non‑negative geometric deconvolution method that scales to millions of single-cell references, improving accuracy and robustness versus existing bulk RNA‑seq deconvolution tools.

• 📰 Patient-derived organoids in functional precision oncology: from experimental models to clinical decision-making
  Amanda Caruso, A. Delvecchio, R. Memeo, M. Lanzino, S. Martinotti — Frontiers in Endocrinology, 2026-07-22
  Reviews patient-derived organoids as functional precision oncology platforms, integrating high-throughput pharmacotyping, co-culture immuno-oncology, and AI analytics for clinically actionable stratification.

• 📰 Liquid Biopsy in Precision Oncology: Clinical Applications and Emerging Roles of Circulating Tumor DNA, Cell-Free DNA, and Extracellular Vesicles
  Z. Kovács, L. Banias, Simona Gurzu — Applied Sciences, 2026-07-22
  Surveys AI-integrated liquid biopsy (ctDNA/cfDNA/exosomal nucleic acids) for diagnosis, MRD, and resistance tracking, highlighting analytical–clinical gaps and ML-driven multimodal fusion opportunities.

• 📰 A 3-dimensional Resnet model for assessment of drug efficacy in 3D cancer models using optical coherence tomography
  Gavrielle R. Untracht, Jan Kaminski, Eike Guldenring, B. S. Nielsen, Kim Holmstrøm, Katrine Hommelhoff Jensen, Peter E. Andersen — PLOS One, 2026-07-24
  Demonstrates that a 3D ResNet on OCT volumes classifies drug response in 3D tumor spheroids with 91.2% accuracy, identifying volumetric biomarkers for high-throughput pharmacologic screening.

🧬 Preprints (arXiv + bioRxiv)

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

• 🧬 NUMonomer enables accurate and scalable nucleic acid structure prediction from primary sequence alone
  si, y.; zhang, s.; chen, l. — bioRxiv, 2026-07-20
  Introduces NUMonomer, an end-to-end deep model that predicts 3D RNA and single-stranded DNA structures directly from sequence without MSAs, matching or surpassing state-of-the-art on CASP16 and long RNAs while being ~100× more efficient.
  Affiliations: UCAS

• 🧬 UniFlow: Unifying protein conformational ensemble generation and machine-learned force fields with a scalable normalizing Flow
  Liu, Y.; Chen, M.; Lin, G. — bioRxiv, 2026-07-20
  Presents UniFlow, an internal-coordinate normalizing flow that unifies protein ensemble generation and coarse-grained machine-learned force fields, enabling fast i.i.d. sampling, exact likelihoods, and stable long-timescale MD within one model.
  Affiliations: Purdue University

• 🧬 X-PAIR: an ultrafast multitask framework for proteome-scale reconstruction of PPI networks and partner-specific interfaces from sequence
  Rescalli, S.; Carbone, A. — bioRxiv, 2026-07-23
  X-PAIR jointly predicts whether two proteins interact and their partner-specific interfaces from sequence using pLM embeddings and lightweight cross-attention, delivering proteome-scale throughput and superior interface localization without MSAs or templates.
  Affiliations: Sorbonne University

• 📄 TriGlue: a Biology-Inspired Generative Model for Generating Molecular Glue-Induced Ternary Complex
  Yuliang Yan, Shuo Yan, Haochun Tang, Yiqin Sun, Enyan Dai — arXiv, 2026-07-24
  TriGlue formulates molecular glue discovery as ternary complex generation with an SE(3)-equivariant interface estimator and interface-conditioned flow matching, jointly generating ligands and rigid-body assemblies for E3–target ternary complexes.

• 🧬 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, 2026-07-21
  Combines RFdiffusion and ProteinMPNN to design de novo integrated domains that reprogram plant NLR immune receptors to recognize new pathogen effectors, validating structure, binding, and immune signaling in planta.
  Affiliations: Centre for Programmable Biological Matter, Department of Biosciences

• 🧬 Activity and specificity trade-offs in adenine base editors
  Lukarska, M.; Oltrogge, L. M.; Nisonoff, H.; Long, Y.; ...; Aquino, C.; Kim, S. E.; Listgarten, J.; Savage, D. F. — bioRxiv, 2026-07-22
  Uses ML-guided library design and high-throughput screens to map adenine base editors onto a single intrinsic deaminase-activity axis, quantifying an activity–specificity trade-off that constrains next-generation ABE engineering.
  Affiliations: University of California, Berkeley

• 🧬 FARM: Forecasting Antibiotic Resistance in Mycobacterium tuberculosis using biophysics and machine learning
  Tasmin, M.; Barethiya, S.; Wang, Y.; Kang, L.; Chen, J. G.; Green, A. G. — bioRxiv, 2026-07-25
  FARM integrates structural, biophysical, protein language model, and physicochemical features to forecast resistance from Mycobacterium tuberculosis mutations, achieving strong temporal generalization and prioritizing hundreds of candidate resistance variants.
  Affiliations: University of Massachusetts

• 📄 Boltzmann-Expected Molecular Design with Decoupled Annealing Flows
  Selma Moqvist, Richard Beckmann, Ross Irwin, Rocío Mercado, Simon Olsson — arXiv, 2026-07-21
  DECAF reframes 3D molecular design around Boltzmann-expected ensemble properties using decoupled annealing flows for graphs and coordinates, enabling objective-agnostic optimization of ensemble means and higher moments without retraining.

• 📄 Antigen-specific Antibody Multi-modal Foundation Model for Functional Antibody Design
  Xiaoliang Shi, Zichen Wang, Runze Ma, Zhongyue Zhang, Shuangjia Zheng — arXiv, 2026-07-22
  AAMFM is an antigen-conditioned multimodal foundation model that learns unified antibody sequence–structure representations with cross-modal adapters and Cal-DPO, improving antigen-specific antibody design and functional relevance.

• 📄 Hypothesis-and-Refinement Learning of Organic Structures from Multimodal Spectroscopic Data
  Chengchun Liu, Zhiyuan Yan, Li Yuan, Hao Li, Boxuan Zhao, Yonghong Tian, Bartosz A. Grzybowski, Fanyang Mo — arXiv, 2026-07-22
  Introduces SpectroMol and MS-Mol2Mol, a multimodal hypothesis–refinement pipeline that fuses rich NMR modalities with a mass-constrained generative prior trained on 400M molecules, achieving 93.8% top-1 spectrum-to-structure recovery and adapting to experimental spectra.

• 📄 PertReason: A Knowledge-Grounded Benchmark and Framework for Cell-State-Conditioned Mechanistic Reasoning of Perturbation Effects
  Dongkwan Kim, Yiming Gao, Yining Yang, Yang Shen — arXiv, 2026-07-21
  PertReason provides a knowledge-grounded benchmark and framework for cell-state–conditioned mechanistic reasoning about genetic/chemical perturbations, exposing failures in faithful reasoning and introducing a model aligned to context-specific pathways.

• 🧬 Deep interpretable learning of sample representations for characterizing disease states in single-cell transcriptomics
  Wagle, M. M.; Wang, Y.; Samanta, S.; Liu, Z.; Patrick, E.; Yang, P.; Kellis, M. — bioRxiv, 2026-07-22
  Phenoverse learns interpretable sample-level disease embeddings from scRNA-seq via cell-type–aware residual encoding, prototype learning, and Perceiver aggregation, predicting severity and revealing reproducible cell-type programs across cohorts.
  Affiliations: MIT CSAIL, The Broad Institute of MIT and Harvard

• 📄 Causal dictionary learning reveals and validates transcription-factor binding features in genomic language models
  Sarwan Ali — arXiv, 2026-07-21
  Combines sparse dictionary learning with causal ablation to extract and validate monosemantic TF-binding features from genomic language models, disambiguating motif signals from GC/repeats and demonstrating causal use for cell-type–specific binding.

• 📄 Do Language Models Dream of Binding Molecules? Benchmarking LLMs under Spatial Constraints
  Thomas MacDougall, Maksim Kuznetsov, Roman Schutski, Rim Shayakhmetov, Maxim Malkov, Vladimir Aladinskiy, Alex Aliper, Alex Zhavoronkov — arXiv, 2026-07-20
  Introduces 3D-Fit to benchmark general-purpose LLMs on pocket-conditioned ligand generation under multiple spatial constraints, revealing emerging multi-constraint handling yet clear gaps to specialized diffusion baselines in SBDD.

• 📄 ABOPD: Antibody CDR Design via On-Policy Distillation
  Zhuo Yang, Jiaying He, Jiaqing Xie, Daolang Wang, Xipeng Qiu, Yuxin Wang, Tianfan Fu, Beilun Wang — arXiv, 2026-07-21
  ABOPD applies on-policy distillation that supervises denoising trajectories with privileged native geometry for antibody CDRs, substantially improving CDR-H3 backbone recovery over standard fine-tuning and offline distillation.

• 🧬 Generative Machine Learning and Microfluidics uHTS: An Efficient Partnership for Enzyme Engineering
  Nair, P. M.; Steinberg, D. M.; Resende, T.; Suarez, A. F.; ...; Oh, V.; Tan, S. H.; Speight, R. E.; Vahidi, A. K. — bioRxiv, 2026-07-24
  Pairs microfluidic uHTS (>30k sequence–function pairs) with a task-specific generative protein model (VSD) to engineer UPO specificity, producing variants with improved desired-product enrichment beyond screen-only baselines.
  Affiliations: Allozymes Pte Ltd

• 📄 Multi-modal transformer for signal classification in nanopore blockade experiments
  Sandro Kuppel, Julian Hoßbach, Samuel Tovey, Christian Holm — arXiv, 2026-07-22
  A multi-modal transformer jointly ingests raw nanopore currents, wavelet images, and engineered features to classify peptides and amino acids, surpassing prior methods and highlighting complementary signal representations.

• 🧬 CHIMIYA-1: An Autoselection Foundation Model for ADMET Property Prediction, Rigorously Benchmarked Against the Therapeutics Data Commons ADMET Group
  Varghese, R.; Tiwary, P.; Oswal, K. — bioRxiv, 2026-07-23
  CHIMIYA-1 is an autoselection foundation model rigorously evaluated on the full TDC ADMET benchmark with multi-run statistics and overlap audits, achieving top-decile performance and leading scores on several endpoints.
  Affiliations: Department of Pharmaceutical Sciences, Philadelphia College of Pharmacy

• 📄 Evolution-Aware MSA Reasoning for Subsampling via Factor Graphs
  Zhangzhi Xiong, Minzhang Li, Haotian Yu, Sixian Shen, ..., Mingrui Li, Jie Zheng, Kewei Tu, Jingyi Yu — arXiv, 2026-07-24
  AP-REASONER casts MSA subsampling as a controllable factor-graph optimization with affinity propagation, outperforming heuristic subsamplers on contact and conformational ensemble prediction while tuning identity/diversity trade-offs.

• 🧬 An automated platform for spatial functional modeling and fingerprint analysis of tissue molecular landscapes
  Hajihosseini, M.; Patino-Martinez, E.; Ghosal, R.; Kaplan, M. J.; Pyne, S. — bioRxiv, 2026-07-21
  SFinx integrates deconvolution, pathway activity reconstruction, and spatial functional data analysis to map localized pathway landscapes and spatially varying pathway–phenotype interactions in spatial transcriptomics.
  Affiliations: Health Analytics Network, LLC