ClaMSA

ClaMSA classifies multiple sequence alignments (MSAs) using phylogenetic models and machine learning to distinguish coding from non-coding codon sequences.

Key Features:

• Input Requirements: Requires a phylogenetic tree and an MSA as input.
• Output: Outputs probabilities indicating the likelihood that an input MSA belongs to specific classes.
• Machine Learning Framework: Incorporates a continuous-time Markov chain (CTMC) machine learning layer trained end-to-end with recurrent neural networks.
• Discriminative Training: Employs discriminative training optimized for classifying coding versus non-coding regions.

Scientific Applications:

• Classification of aligned codon sequences: Distinguishes coding and non-coding regions in codon-aligned MSAs.
• Comparative genomics benchmarking: Applied to vertebrate and fly alignments, reporting six times fewer false positives than existing methods at an equivalent true positive rate.

Methodology:

Integration of a continuous-time Markov chain (CTMC) machine learning layer with recurrent neural networks trained end-to-end using discriminative training optimized for classifying coding versus non-coding regions on phylogenetic-tree-aware MSAs.