GenixRL Scores Database: Pre-computed Variant Pathogenicity Scores from dbNSFP (BayesDel, ClinPred, MetaRNN)

This dataset provides pre-computed pathogenicity scores for genomic variants, specifically designed for use with the GenixRL (https://github.com/shabbas313/GenixRL). The database is derived from dbNSFP (dbNSFP5.2a_grch38.gz) and includes scores from key predictors: BayesDel_noAF, BayesDel_addAF, ClinPred, and MetaRNN. Dataset Contents genixrl_scores_db.tsv.gz: Compressed TSV file containing variant annotations with columns for chromosome, position, reference allele, alternate allele, and scores (BayesDel_noAF_score, BayesDel_addAF_score, ClinPred_score, MetaRNN_score). The file is tab-separated and gzip-compressed for efficiency. genixrl_scores_db.tsv.gz.tbi: Tabix index file for fast querying of the TSV using tools like pysam in Python. Methods and GenerationThe database was generated by querying dbNSFP (a comprehensive database of non-synonymous functional predictions) for a curated set of variants relevant to the GenixRL model. Scores were extracted for the following predictors: BayesDel_noAF and BayesDel_addAF: Bayesian deleteriousness scores for missense variants. ClinPred: Clinical predictor score for pathogenic potential. MetaRNN: Meta-ensemble RNN score for variant pathogenicity. Normalization is applied in the GenixRL model: BayesDel scores are scaled to [0,1] using min=-1.29334 and max=0.75731. Missing values are imputed using training medians from the associated model artifacts.