transformer / foundation model

LucaCell is a sequence-centric foundation model that improves cross-species single-cell analysis by using mRNA sequence embeddings instead of fixed gene identifiers.

V-TRACE provides an end-to-end platform for automated animal behavior analysis from video, achieving high accuracy and throughput across species and contexts.

Inferring tested-but-undetected pairs from prey lists of 5,530 published interaction screens yields 97 million tested human protein pairs and 18,982,508 non-interactions (≥3 tests) for calibrating screens and predictors.

ABCP_finder predicts anti-breast cancer peptides using transformer embeddings, achieving high accuracy and supporting biological relevance through external validation.

GEM-GPT generates personalized therapeutic molecules to shift cell-type-specific disease states toward healthy phenotypes, outperforming baselines in chemical perturbation and CRISPR experiments.

This paper introduces an environment-aware DNA language model to prioritize genomic regions associated with stress response in maize using stress-specific prompts and transformer-based modeling.

HyphAeon is a transformer model that accelerates detection of episodic diversifying selection in mammalian genomes and enables new analytical capabilities beyond traditional dN/dS methods.

MycoAI ITS classifiers trail alignment, SINTAX and HiTaC on full-length ITS and drop to 28.3% and 18.5% seen-genus family accuracy on ITS2, as predictions track flanking regions.

This paper compares segmentation methods for plant phenotyping, finding that pretrained vision transformers outperform color thresholding and supervised U-Nets in generalization across species and conditions.

This paper evaluates a self-architecting transformer for protein representation, showing it achieves better performance than existing models with a smaller dataset.

A conditional latent diffusion transformer predicts immunofluorescence images of 12,800 human proteins from nucleus, ER and microtubule stains, outperforming PUPS and CELL-Diff, including on an unseen drug-treated cell line.

The paper introduces Atlas, a microbiome dataset, and Waypoint, a series of transformer-based models, showing improvements in downstream tasks through pretraining.

This paper introduces SHARP, a method to enhance Hi-C data resolution while avoiding artificial structures by decomposing signals and using multi-scale attention.

The paper develops deep learning models to detect pain decreases using physiological signals during fluctuating noxious heat stimulation, achieving 0.854 AUROC and 5.75s latency.

This paper introduces InstaNovo-FM, a self-supervised foundation model for proteomics that learns from tandem mass spectra to enable transferable applications across datasets and instruments.

The paper uses AI to quantify cysteine in pea cultivars from SERS spectra, finding that a 1D-CNN outperforms other models in accuracy and generalization.

The paper introduces CiRF-ViT, a vision transformer with per-neuron locality priors, achieving high performance in small-scale neural prediction tasks.

This study uses ESM3 to generate knotted proteins, finding high success rates with guided decoding and robustness in real knotted proteins.

The paper introduces AtlasFold, a protein structure prediction method using metagenomic-scale language models, achieving state-of-the-art accuracy and efficient inference.

This study shows that JEV infection in pig fetuses leads to genetic diversity in the virus and transcriptional changes in offspring, suggesting potential for onward transmission and long-term health impacts.

The paper shows a sequence-only pipeline using TabPFN achieves high accuracy in multi-activity antimicrobial peptide profiling, outperforming complex deep models.

PhomiNeuro optimizes NIR neuromodulation light fluence using a differentiable surrogate model, improving speed and accuracy for personalized treatment planning.

The paper evaluates text-guided segmentation using SAM3 for weakly supervised leaf symptom segmentation, finding text guidance matches or outperforms conventional methods but is prompt-sensitive, with hybrid approaches improving robustness.

The paper screens variants of a Form II Rubisco to improve oxygen tolerance, identifies key mutations, and uses a transformer model to guide engineering.

This paper develops a machine-vision workflow to convert video footage of salmonids into individual encounter histories for capture-recapture analysis, achieving high accuracy in fish detection and re-identification.

CatESO optimizes enzyme sequences to improve catalytic turnover by integrating kcat prediction with differentiable design, achieving higher predicted activity and structural confidence than existing methods.

The paper identifies a novel phage resistance mechanism in Pseudomonas putida through experimental evolution, revealing a secreted protein Psh that confers broad-spectrum phage resistance without fitness costs.

A set-aware latent transport model predicts treated cell populations from unpaired controls, keeps perturbation-specific signal in CRISPR, drug and cytokine data, and guided a cytokine choice tested in three donors.

scMaize provides a maize-specific foundation model and integrated single-cell atlas, showing improved classification and cross-species generalizability with functional priors.

Across 13 single-cell foundation models and 94 datasets, PCA and established methods usually match or beat them; pretrained embeddings help mainly with scarce labels, rare cells and open-set detection.