Splice-Site Prediction from Genomic Sequences

A 1D convolutional neural network with self-attention reads 401 bp of human DNA around every GT and AG dinucleotide and predicts whether it is a splice donor, a splice acceptor or neither. Longer sequences, up to whole pre-mRNAs, are scanned candidate by candidate and the calls are paired into candidate introns. Splice sites come from the Ensembl release 116 MANE Select annotation and the genome sequence from NCBI (GRCh38). Four numeric signal encodings of DNA are compared, and the model is tested on whole genes and on ClinVar variants. A complete stand-alone MATLAB version covers signal encoding and prediction.

Category: Bioinformatics - Genomics and Sequence Analysis (Deep Learning, 1D-CNN, Attention, MATLAB, Python)

Bioinformatics Project B09 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.