Computational Modeling and Optimization of Lipase Production by Bacillus Species: An In Silico Study

Abstract

Microbial lipases are among the most industrially versatile biocatalysts, with extensive applications in detergent formulation, food processing, and biodiesel synthesis, and members of the genus Bacillus are among their most widely studied bacterial producers. This study presents a literature-derived, fully reproducible computational analysis of Bacilluslipase production, constructed entirely from peer-reviewed experimental data compiled from five independent published studies spanning four species/strains (Bacillus subtilis TTP-06 and NS8, Bacillus aryabhattai SE3-PB, Bacillus amyloliquefaciens, and Bacillus holotolerans VSH09). A structured dataset of 21 data points was assembled across six physicochemical variables — temperature, pH, incubation time, carbon-source concentration, nitrogen-source concentration, and agitation speed — with unreported values retained explicitly as "Not Reported" rather than estimated or imputed. Correlation analysis across the pooled dataset identified agitation speed (r = 0.573, p = 0.013) and pH (r = 0.518, p = 0.023) as the strongest determinants of reported lipase activity, while Principal Component Analysis revealed that the dataset's structure was heavily shaped by a single dominant source study. Within the only subset supporting full multivariate modeling (n = 15, from a Box–Behnken design), Multiple Linear Regression, Response Surface Methodology, Random Forest Regression, Support Vector Regression, and an Artificial Neural Network were compared under leave-one-out cross-validation; Random Forest achieved the best performance (R² = 0.261, RMSE = 3.554 U/mL), while quadratic Response Surface Methodology identified a statistically significant, concave carbon-source effect and an optimum of 2.74% molasses and 1.40% peptone, yielding 27.86 U/mL. These findings support carbon-source concentration as a priority optimization target for Bacillus lipase production while highlighting the current scarcity of harmonized, multi-variable literature data needed for broadly generalizable computational optimization models.