Anna G. Green

The paper predicts antibiotic resistance mutations in Mycobacterium tuberculosis by integrating structural context with mutation data, finding that structural clustering improves prediction accuracy.

The paper introduces a sparse feature circuit framework to identify causal features in ESM-2 for zero-shot mutation effect prediction, showing that shared circuits explain mutation effects through 3D contact residues.

The paper develops FARM, a framework to predict antibiotic resistance in Mycobacterium tuberculosis by combining biophysical and machine learning features, achieving high accuracy in distinguishing resistance-associated variants.

AMR-GNN predicts antimicrobial resistance using a graph neural network framework that integrates multiple genomic representations from whole-genome sequencing data.

The paper identifies a frameshift mutation in rlmN as a significant contributor to linezolid resistance in Mycobacterium tuberculosis, beyond known mutations in rrl and rplC.

BIG-TB evaluates sequence-based machine learning models for predicting antibiotic resistance and variant attribution using *Mycobacterium tuberculosis* genomes, showing DNA-based models outperform protein-based ones.

The paper identifies key latent features in protein language models that drive contact prediction through motif and domain recognition using causal analysis.

The paper identifies chemical features affecting mycomembrane permeation and antitubercular activity using a bioorthogonal assay and machine learning.

This paper presents a deep convolutional neural network to predict antibiotic resistance in M. tuberculosis isolates, achieving state-of-the-art accuracy and enabling interpretable feature analysis.

The paper identifies genome-wide mutational dependencies in Mycobacterium tuberculosis evolution, showing how mutations interact to confer antibiotic resistance and detect environmental pressures.

The paper uses sequence coevolution to predict protein interactions in E. coli, identifying 620 new interactions and their structures.

The paper introduces EVcouplings, a Python framework for coevolutionary sequence analysis that enables structure prediction and mutation effect assessment.

The paper determines the structure of human PcdhYB3's antiparallel EC1-4 homodimer and shows that subfamily-specific specificity arises from loop region diversification, while EC1/EC4 interfaces use hydrophobic interactions for non-selective dimerization.

This paper uses sequence co-evolution to predict 3D contacts and structures of protein complexes, achieving accurate results on known structures and predicting contacts for unknown ones.