ML Models

Revised M4 and M5 models

Separate model versions with corrected inputs. M4 scores annotation classes; M5 scores source-specific translation-efficiency aggregates. Neither score establishes therapeutic efficacy.

Read fresh holdout results, baselines and limitations

M4 transcript geometry v2

Provide exon, exons, cds and stops as genomic coordinate pairs, plus strand (+ or −). Coordinates are one-based and inclusive. The candidate must belong to this transcript; all three stop-codon bases must be annotated. The model computes corrected features for you.

M5 sequence context

Segment lengths must sum to sequence length. A canonical AUG must start the CDS. Folding is included in this request and can take longer than the inference-only benchmark.