Simone Marini

@simonemarini.bsky.social

2446 Followers 750 Following

AI, BioML, compbio, data science | pathogens, AMR antimicrobial and antibiotic res, inflammation, CBD, single-cell RNA seq, metagenomics | Asst Prof @ University of Florida, AI advisor @ enGenome | Prev: UMich, UniPV, KyotoU | Immigrant 🇪🇺🇭🇰🇯🇵🇺🇸

Gaussian kernel density estimate plots show that the observed distribution of variants across positions in the SARS-CoV-2 S1 subunit both pre- (training) and post-February 2021 (test) are well approximated by SARITA.

Boxplot displaying the distribution of mutation false mutation rate (FMR) across synthetic sequences. The FMR is calculated as the proportion of additional mutations—those not present in the best-aligned sequence—relative to the total mutations generated by the model. SARITA shows superior performance against competing models.

Percentage of generated sequences with LD < 4, 10, 100, 500, 700, 1500, and 2000 against the best aligned test sequence. All SARITA models achieve LD<4 with at least 31.7% of the results.

Boxplot showing the PAM30 distribution scores computed on the synthetic sequences (higher PAM30 score means that the sequences are strongly related). SARITA models outperform the others.