CN-RNN: a deep learning framework for copy number variants detection with exome sequencing data
Tool / method
Deep learning classification of candidate CNVs: BiLSTM on read depth and MLP on GC content, mappability and exon length
Summary
CNV detection from exome data remains limited by high false-positive rates and low sensitivity for short CNVs. CN-RNN works in two stages: high-sensitivity candidate generation, then deep learning classification combining a BiLSTM branch, which captures depth changes and dependencies across neighbouring exons, with an MLP branch that incorporates region-level features (GC content, mappability, exon length). The model was trained on parent-offspring trios from the Autism Sequencing Consortium, with labels built through Mendelian inheritance filtering. In internal validation and on two external datasets built with distinct reference sets, CN-RNN consistently outperformed existing exome CNV callers and deep learning methods. The code is available on GitHub.
Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.
Analysis
The need is real for any laboratory reporting exomes, since short CNVs are the weak link, as the abstract points out. But no sensitivity or precision figures are given and the outperformed tools are not named, which makes the size of the gain impossible to judge. CN-RNN still needs to be tested on confirmed pathogenic CNVs in a diagnostic setting, beyond the Mendelian labels of a research cohort.
Analysis by Dr Thibaut Benquey
Why this score?
Clinical impact: 2/3 · Evidence strength: 2/3 · Novelty: 1/2 · Sample size: 1/1 · Publication status: 1/1 → Total: 7/10
Keywords
Every Wednesday · Annotated selection · Free · Unsubscribe anytime