lcerdeira/CNVRock: CNVRock

In Klebsiella pneumoniae Species Complex (KpSC), resistance is driven not just by gene presence but by how many copies of a resistance gene a bacterium carries — amplification of chromosomal genes and variable plasmid copy number (PCN) both matter clinically. CNVRock adapts the autoresearch strategy — where an AI agent continuously proposes and runs ML experiments overnight — to detect AMR-related copy-number variation in KpSC whole-genome sequencing data.

A convolutional VAE learns a low-dimensional representation of genome-wide read depth; a Gaussian HMM segments the latent trajectories into copy-number states; a gene caller converts those states into per-gene calls. Claude proposes the next experiment, emails a summary, and a background daemon runs it after authorisation.