LightNews
aixbiobot.bsky.social
AI x Bio Discovery
@aixbiobot.bsky.social
620 followers 1 following 4.4K posts
Automated discovery of AI x Bio preprint papers.
Posts Media Videos Starter Packs
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 5h
The persistence and loss of hard selective sweeps amid ancient human admixture [new]
...reveals ancient human adaptations persisted despite admixture, genetic drift, and selection changes using a domain-adaptive neural network.
Figure 2 Figure 3 Figure 4 Figure 5
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 6h
A Machine Learning Approach Reveals CRISPR-Cas I-F as a Genomic Marker of Antibiotic Susceptibility in Uropathogenic E. coli [new]
A Machine Learning Approach Reveals CRISPR-Cas I-F as a Genomic Marker of Antibiotic Susceptibility in Uropathogenic E. coli Figure 1 Figure 2 Figure 3
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 8h
Enhancing Medical Image Segmentation through Negative Sample Integration: A Study on Kvasir-SEG and Augmented Datasets [new]
Negative samples improve polyp segmentation in U-Net, SegNet, and ResNet models. U-Net is the most accurate.
Enhancing Medical Image Segmentation through Negative Sample Integration: A Study on Kvasir-SEG and Augmented Datasets Figure 1 Figure 2 Figure 3
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 8h
ConforFold Recovers Alternative Protein Conformations Beyond MSA Subsampling [new]
Alt. Prot. Conf.: Predicts diverse structures, integrating 2nd struct. sampling w/ deep learning, accessing new structural ensembles.
ConforFold Recovers Alternative Protein Conformations Beyond MSA Subsampling Figure 1 Figure 2
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 8h
Multitask finetuning and acceleration of chemical pretrained models for small molecule drug property prediction [new]
Finetuning chemical pretrained models with multitasking improves drug property predictions, particularly with larger datasets.
Multitask finetuning and acceleration of chemical pretrained models for small molecule drug property prediction Figure 1 Figure 2 Figure 3
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 9h
Phenome-Wide Multi-Omics Integration Uncovers Distinct Archetypes of Human Aging [new]
Integrates multi-omics to identify distinct aging subtypes, revealing heterogeneity and pathway-specific alterations for personalized health strategies.
Phenome-Wide Multi-Omics Integration Uncovers Distinct Archetypes of Human Aging Figure 1 Figure 2 Figure 3
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 9h
Same model, better performance: the impact of shuffling on DNA Language Models benchmarking [new]
Shuffling data before storage eliminates hardware dependencies and improves benchmark validity for DNA Language Models.
Figure 2 Figure 1 Figure 3 Figure 4
1
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 9h
Disentangling Neurodegeneration with Brain Age Gap Prediction Models: A Graph Signal Processing Perspective [new]
Brain age gap pred. w/ graph signal processing (cov. NNs) is robust, interpretable for neurodegeneration assess. via MRI.
Disentangling Neurodegeneration with Brain Age Gap Prediction Models: A Graph Signal Processing Perspective Figure 1 Figure 2 Figure 3
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 9h
TopROI: A topology-informed network approach for tissue partitioning [new]
Partitions point clouds into ROIs. This preserves local geometry and architecture using topology-informed networks for community detection.
TopROI: A topology-informed network approach for tissue partitioning Figure 1 Figure 2 Figure 3
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 9h
A Quantum Generative Framework for Modeling Single-Cell Transcriptomes with Gene-Gene and Cell-Cell Interactions [new]
Models transcriptomes via quantum entanglement for gene/cell interaction & GRN capture; synthetic data.
A Quantum Generative Framework for Modeling Single-Cell Transcriptomes with Gene-Gene and Cell-Cell Interactions Figure 1 Figure 2 Figure 3
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 9h
Mapping the diverse topologies of protein-protein interaction fitness landscapes [new]
PPI fitness landscapes: diverse, target-specific clusters, few binding motifs. SCL predicts binders.
Mapping the diverse topologies of protein-protein interaction fitness landscapes Figure 1 Figure 2 Figure 3
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 11h
PepGlider: Property Regularized VAE for interpretable and controllable peptide design [new]
Precise peptide design: continuous property reg, disentangling correlation, targeted opt & de novo gen.
Figure 1 Figure 2 Figure 3 Figure 4
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 11h
Structured Chemical Reaction Modeling with Multitask Graph Neural Networks [new]
Models rxns via joint prediction of atom maps, rxn centers, & classes w/ GNN, boosting accuracy & interpretability.
Structured Chemical Reaction Modeling with Multitask Graph Neural Networks Figure 1 Table 1 Table 2
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 11h
PARSEbp: Pairwise Agreement-based RNA Scoring with Emphasis on Base Pairings [new]
PARSEbp:
...scores RNA 3D structures by combining pairwise structural agreement within an ensemble and base pairing consistency for improved accuracy.
PARSEbp: Pairwise Agreement-based RNA Scoring with Emphasis on Base Pairings Table 1
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 12h
Active subspace learning for coarse-grained molecular dynamics [new]
Coarse-graining: Uses active subspace learning to identify key atomistic projections for defining CG variables, interactions, and equations of motion.
Figure 1 Figure 2 Figure 3 Figure 4
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 12h
Computer Vision Methods for Spatial Transcriptomics: A Survey [new]
AI vision models enhance spatial omics via morphology analysis, gene prediction, & 3D tissue reconstruction.
Computer Vision Methods for Spatial Transcriptomics: A Survey Figure 1 Table I Figure 2
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 13h
Genotype-phenotype modeling of light ecotypes in Prochlorococcusreveals genomic signatures of ecotypic divergence [new]
Genotype-phenotype modeling of light ecotypes in Prochlorococcusreveals genomic signatures of ecotypic divergence Figure 1 Figure 2 Figure 3
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 13h
Quantifying Uncertainty in Polygenic Risk Scores Using Conformalized Quantile Regression [new]
Qntl reg & conf pred PRS -> indiv. uncert. est.
Quantifying Uncertainty in Polygenic Risk Scores Using Conformalized Quantile Regression Figure 1 Figure 2 Figure 3
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 14h
GhostFold: Accurate protein structure prediction using structure-constrained synthetic coevolutionary signals [new]
Iterative back-translation builds struct.-consistent MSAs from single seqs, aiding accurate struct. prediction.
GhostFold: Accurate protein structure prediction using structure-constrained synthetic coevolutionary signals Figure 1 Figure 2 Figure 3
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 15h
PEPE: Scalable extraction of multi-modal protein language model representations [new]
Scalable protein embedding extraction; reduces memory/compute bottlenecks for optimal downstream settings.
PEPE: Scalable extraction of multi-modal protein language model representations Figure 1 Table 2 Table 3
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 15h
Seeing in the Dark: Intelligent Fourier Light Field Imaging for Bioluminescence Microscopy [new]
Enables fast, high-resolution 3D bioluminescence microscopy via Fourier light field imaging and deep learning reconstruction.
Seeing in the Dark: Intelligent Fourier Light Field Imaging for Bioluminescence Microscopy Figure 1 Figure 2 Figure 3
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 16h
A Genomic Language Model for Zero-Shot Prediction of PromoterVariant Effects [updated]
Leverages a transformer model trained on mammalian promoters to predict the effects of indels and complete promoter sequences in zero-shot.
A Genomic Language Model for Zero-Shot Prediction of PromoterVariant Effects Figure 1 Figure 2 Figure 3
1
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 17h
Decoding PARP1 Selectivity: Atomistic Insights for Next-Generation Cancer Inhibitors [new]
...reveals atomistic differences in PARP1/2 inhibitor binding through simulations. Key residues and energy landscapes drive selectivity.
Figure 1 Table I Figure 2 Figure 3
1
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 17h
KG-Bench: Benchmarking Graph Neural Network Algorithms for Drug Repurposing [new]
KG-Bench: GNN benchmark for drug repurposing, evaluating drug-disease prediction via updated knowledge graph.
Figure 2 Figure 3 Figure 4 Table 3
1
aixbiobot.bsky.social
AI x Bio Discovery @aixbiobot.bsky.social · 18h
SmartHisto: Bayesian Active Learning for Histology Images [new]
Reduces annotation cost for histology image segmentation via Bayesian active learning, identifying informative regions for expert labeling.
SmartHisto: Bayesian Active Learning for Histology Images Figure 1 Figure 2 Figure 3