Huazhong University of Science and Technology

EEG foundation model coupling spatial and temporal transformer branches through a shared soft mixture-of-experts, adapted by tuning 5.1% of weights.

Multimodal microbiome foundation model pretrained on 1.8 million samples. Frozen representations transfer to biome classification and forecasting.

Unified bio-language Mixture-of-Experts model spanning DNA, protein sequence and structure, and biological text across eight task families.

Microscopy image restoration foundation model unifying 8 tasks across 5 modalities and 2D/3D data, with zero-shot inference on unseen systems.

Plant genome foundation model pairing a bidirectional Mamba backbone with sparse Mixture-of-Experts, pretrained on 25.4B nucleotides from 42 species.

RNA secondary structure prediction that fuses base-pair priors from four orthogonal folding tools, reaching 0.709 F1 on cross-family bpRNA-new.

Cryo-EM and cryo-ET map enhancement model that sharpens density maps with a Mamba-based dual-branch UNet and local resolution-guided learning.

Protein-protein interaction predictor that adds contact-guided dual attention and a geometric encoder to frozen protein language model embeddings.

Alignment-free biosynthetic gene cluster detection and annotation from ESM-2 gene embeddings in genomic context, up to 102x faster than antiSMASH.

Generative foundation model that co-generates sequence and 3D coordinates for proteins, small molecules, and crystals under functional objectives.

Plant genomic foundation model with a 64 kb single-nucleotide context that predicts gene structures and generates de novo plant gene sequences.

DNA language model pretrained jointly on English, protein, and genomic text under one BPE vocabulary, transferring text segmentation skills to DNA.

Generative microbiome language model writing disease-conditioned community profiles and predicting post-transplant composition for FMT donors.

Microbiome community foundation model pretrained on 263,302 samples, encoding genus abundance as ranked tokens for classification and generation.

Instruction-tuned gene language model extending LLaMA-7B with merged DNA and protein BPE vocabularies to answer sequence tasks as chat prompts.

Peptide-spectrum match rescoring for DDA proteomics, learned end to end from raw MS2 spectra and peptide sequence across 271 million PSMs.

Genomic sequence classification answered through natural-language prompts by one GPT-2 pretrained on mixed DNA and English under one BPE vocabulary.

Medical vision-language model trained on the MedTrinity-25M dataset, answering questions and generating text about radiology and histology images.

Antibody CDR design framework pairing a pretrained antibody language model with a hierarchical graph neural network for one-shot CDR generation.