Robust semi-supervised scRNA-seq integration from virtual adversarial learning [new]
...incorporates marker gene information to preserve fine-grained cell subtype distinctions, robust to noisy or incomplete annotations.
Bridging three-dimensional molecular structures and artificial intelligence with a conformation description language [new]
enables AI models to perform 3D molecular modeling tasks, including conformer prediction and 3D generation.
AI x Bio Discovery
AI x Bio Discovery
Single-allele chromatin folding is organized by recurrent motifs that are reweighted across cell types and states [updated]
from loop extrusion, a limited motif rep. drives diverse struct&gene activity via alt. occup., not new confs.
Combinatorial docking and molecular generation to navigate over 100-billion molecules for prospective ligand discovery [new]
exhaustive screening and rapid navigation of these massive libraries via generative framework for drug disc.
Interpretable enzyme function prediction via sparse autoencoder features of ESMC across the microbial protein universe [new]
…leverages interpretable semantic features to predict/discover microbial dark enzyme functions, reveals mech. relevant concepts.
HoloCell: A Generative Foundation Model for Holistic Cellular Modeling [new]
...integrates single-cell epigenomics, transcriptomics, and proteomics via hierarchical tokenization for unified state mapping and generative simulation.
Pseudoperplexity Probes Memorization in Protein Language Models [new]
...by comparing its values on training vs. novel sequences, revealing pLMs exhibit detectable memorization of their training data.
Folding the unfoldable 2: using AlphaFold and ESMFold to explore spurious proteins [new]
...reveals high-conf structures for short spurious sequences, yet score analysis and a model aid their large-scale identification.
Investigation of In Vivo Silk Scaffold Degradation by Decoupling Tissue Ingrowth Using a GPR-Driven Digital Twin Framework [new]
...decouples polydeg. from tissue ingrowth, enabling robust longterm fate pred of degradable biomats.
Sequence-Based Therapeutic Peptide Classification with Augmented Negative Sampling [new]
employs a CNN trained on comprehensive DB, leveraging Markov model-gen. decoys to classify from sequence and capture functional motifs.
