AI x Bio Discovery

SR2P: an efficient stacking method to predict protein abundance from gene expression in spatial transcriptomics data

Figure 1

Figure 2

Figure 3

SR2P: an efficient stacking method to predict protein abundance from gene expression in spatial transcriptomics data [new]
...predicts prot. abund. from RNA expr., bypassing spatial multi-omics lim. to ID imm. states/markers in tumor

Application of large language models to the annotation of cell lines and mouse strains in genomics data

Figure 1

Table 1

Application of large language models to the annotation of cell lines and mouse strains in genomics data [new]
...assist in identifying & mapping free-text cell line and mouse strain entries to ontologies in genomic metadata curation.

A Modular Framework for Automated Segmentation and Analysis of AFM Imaging of Chromatin Organization

Figure 1

Figure 2

Figure 3

A Modular Framework for Automated Segmentation and Analysis of AFM Imaging of Chromatin Organization [new]
...automates quantifying nanoscale chromatin org. from AFM, revealing prot-spec sigs & enabling label-free nuc. spacing anal.

Popformer: Learning general signatures of positive selection with a self-supervised transformer

Figure 1

Table 1

Figure 2

Popformer: Learning general signatures of positive selection with a self-supervised transformer [new]
encoding general genetic var. patterns through self-supervised pre-training on real human data for broad evolutionary inference.

Abstract NA

Figure 1

Figure 2

Circular RNA identification using a genomic language model and a small number of authenticated examples [new]
achieved by integrating curriculum learning with gLM finetuning, leveraging noisy candidates to overcome limited ver. data.

Minimal Amino Acid Alphabet for Protein Design

Figure 1

Figure 2

Figure 3

Minimal Amino Acid Alphabet for Protein Design [new]
computational design using reduced amino acid alphabets (2-10 AAs) shows structural complexity increases with alphabet size, implying early globular protein formation.

Phenotypic reversion and target prioritization for cellular inflammation via representation learning with foundation models

Figure 1

Figure 2

Figure 3

Phenotypic reversion and target prioritization for cellular inflammation via representation learning with foundation models [new]
ID gene targets shifting inflam. cell profiles to healthy states, enhanced by disease stimuli.

Diffusion-ACP39: A Decoder-Adaptive Latent Diffusion Framework for Generative Anticancer Peptide Discovery

Table 1

Figure 1

Figure 2

Diffusion-ACP39: A Decoder-Adaptive Latent Diffusion Framework for Generative Anticancer Peptide Discovery [new]
...utilizes a synchronized seed autoencoder to generate novel ACPs 5-39 amino acids long.

AI-Driven Generation of Cortisol-Binding Peptides for Non-Invasive Stress Detection

Figure 1

Figure 2

Figure 3

AI-Driven Generation of Cortisol-Binding Peptides for Non-Invasive Stress Detection [new]
...uses generative AI, integrating sequence and structure models, to screen a 10K peptide library and identify high-affinity cortisol binders.

Figure 2

Figure 1

Figure 4

Figure 3

A Zero-Inflated Hierarchical Generalized Transformation Model to Address Non-Normality in Spatially-Informed Cell-Type Deconvolution [updated]

ProtNHF: Neural Hamiltonian Flows for Controllable Protein Sequence Generation

Figure 1

Figure 2

Figure 3

ProtNHF: Neural Hamiltonian Flows for Controllable Protein Sequence Generation [new]
achieved through continuous, inference-time control of properties via additive analytical bias functions, avoiding model retraining.

PAVR: High-Resolution Cellular Imaging via a Physics-Aware Volumetric Reconstruction Framework

Figure 1

PAVR: High-Resolution Cellular Imaging via a Physics-Aware Volumetric Reconstruction Framework [new]
integrates single-shot vol. acq. & fast, end-to-end physics-aware recon., trained in silico for sample-indep. 3D cell imaging.

ESGI: Efficient splitting of generic indices in single-cellsequencing data

Figure 1

Figure 2

Figure 3

ESGI: Efficient splitting of generic indices in single-cellsequencing data [new]
..., a flexible framework for demultiplexing complex, variable-length, & error-prone barcodes (with indels) from diverse single-cell seq designs.

Figure 2

Figure 3

Figure 4

Figure 5

Evolutionary algorithms accelerate de novo design of potent Nectin-4-specific cancer biologics [new]
...via AI-GA integration to efficiently explore sequence-structure space, enabling functional biologics for challenging targets.

Single-Cell Genomics Decontamination with CellSweep

Figure 1

Figure 2

Figure 3

Single-Cell Genomics Decontamination with CellSweep [new]
removes ambient molecules and global bulk contamination, clarifying molecular profiles and cellular identities.

A latent space thermodynamic model of cell differentiation

Figure 1

Figure 2

Figure 3

A latent space thermodynamic model of cell differentiation [new]
models differentiation on a Waddington landscape, inferring cell state, dev. flow, & cell plasticity reconstruct trajectories, predict fate, & reveal regulator effects.

Reliable prediction of short linear motifs in the human proteome

Figure 1

Figure 2

Figure 3

Reliable prediction of short linear motifs in the human proteome [new]
...is achieved via deep learning using refined data and protein embeddings to identify novel motifs and precise protein-protein interactions.

Figure 1

Figure 2

Table 2

Table 3

What Do Biological Foundation Models Compute? Sparse Autoencoders from Feature Recovery to Mechanistic Interpretability [new]
Model activations reveal consistent bio features x-scale; but exp validation needed for learned mechanisms.

Automated Cell Type Annotation with Reference Cluster Mapping [updated]
...by combining optimal transport and integer programming to precisely map scRNA clusters to established reference datasets across technologies, tissues, and species.

An Energy Landscape Approach to Miniaturizing Enzymes using Protein Language Model Embeddings

Figure 1

Figure 2

Figure 3

An Energy Landscape Approach to Miniaturizing Enzymes using Protein Language Model Embeddings [new]
identifies compact seqs retaining catalytic site struct by sampling PLM-informed E-landscape, validated w/ structure prediction & MD.

bMAE: Masked Autoencoder Latent Representations for Bulk RNA-seq Tissues

Figure 1

Figure 2

Figure 3

bMAE: Masked Autoencoder Latent Representations for Bulk RNA-seq Tissues [new]
learns compressed, tissue-discriminative latent spaces, generalizing to unseen tissues and revealing multi-scale hierarchical structure.

Automated high-throughput fabrication of patient-specific vessel-on-chips enables a generative AI digital twin--Cascade Learner of Thrombosis (CLoT) for personalized thrombosis prediction

Figure 1

Figure 2

Figure 3

Automated high-throughput fabrication of patient-specific vessel-on-chips enables a generative AI digital twin--Cascade Learner of Thrombosis (CLoT) for personalized thrombosis prediction [new]

Functional Locality-Aligned Learning Reveals Structure-Function Causality in Enzyme Kinetics

Figure 1

Figure 2

Figure 3

Functional Locality-Aligned Learning Reveals Structure-Function Causality in Enzyme Kinetics [new]
Aligning inductive biases w/ local struct determinants, prioritize catalytic pockets & integrating substr 3D geom for mech. insights.

Miniprotein inhibitors of the Staphylococcus aureus efflux transporter NorA

Figure 1

Figure 2

Figure 3

Miniprotein inhibitors of the Staphylococcus aureus efflux transporter NorA [new]
...were designed and validated by cryo-EM to bind NorA's substrate pocket, blocking drug efflux.

Perturbation-guided mapping of colorectal cancer cell states to causal mechanisms

Figure 1

Figure 2

Figure 3

Perturbation-guided mapping of colorectal cancer cell states to causal mechanisms [new]
Mapping IDs distinct malignant states & links them to perturbations & therapeutic responses via integrated observational & perturbation atlases.

Explainable Physicochemical Determinants of Protein Ligand Binding via Non-Covalent Interactions

Figure 1

Figure 2

Figure 3

Explainable Physicochemical Determinants of Protein Ligand Binding via Non-Covalent Interactions [new]
Predicts binding likelihood & learns residue int. pat. from phys. non-cov. interactions, using seq data & sup. interaction maps.

Figure 1

Figure 2

Figure 3

Figure 4

Massive-scale single-nucleus multi-omics identifies novel rare noncoding drivers of Parkinson's disease [new]
...leveraging multi-omic data (millions nuclei) & ML to pred var effects on gene regul, linking to sporadic/familial forms.

Genome-wide classification of tumor-derived reads from bulk long-read sequencing

Figure 1

Figure 2

Figure 3

Genome-wide classification of tumor-derived reads from bulk long-read sequencing [new]
...is achieved using a transformer model that classifies individual reads based on their methylation patterns, trained by somatic mutations.

Figure 1

Figure 2

Figure 3

Figure 4

FoldARE, an RNA secondary structure analysis and prediction tool via generative pseudo-SHAPE modeling [new]
extracts single-strandedness from in silico ensembles to create pseudoSHAPE constraints for prediction and ensemble analysis.

Figure 1

Figure 2

Figure 3

Figure 4

FoldARE, an RNA secondary structure analysis and prediction tool via generative pseudo-SHAPE modeling [new]
uses pseudo-SHAPE from in silico structural ensembles to guide SHAPE-compatible folding algorithms for RNA struct prediction.