TITLE:
QSAR Models: Exploring Limits in Three Cases
AUTHORS:
El Hadji Sawaliho Bamba
KEYWORDS:
Multiple Linear Regression, QSAR, Tetrahydroisoquinoline, Arylidene Rhodanine, Thiazoline, Thiazine
JOURNAL NAME:
Computational Chemistry,
Vol.13 No.3,
June
25,
2025
ABSTRACT: This article critically assessed the validity of five multiple linear regression models across three separate studies. The first examined the cytotoxic properties of N-tosyl-1,2,3,4-tetrahydroisoquinoline compounds. The second evaluated the antiproliferative effects of 1,3,5-arylidene rhodanines. The last explored the antitumour potential of thiazoline or thiazine derivatives. Despite limited sample sizes, the model validation showed robust performance and predictive capabilities. However, their forecasts lacked accuracy. The authors validated their models by assessing the fit training data and generalization ability. The gaps weren’t clearly defined, and outliers were only partially considered. The cytotoxicity study of N-Tosyl-1,2,3,4-Tetrahydroisoquinoline used a ±2 standardized residual. Non-random sampling can introduce selection bias. Ignoring dispersion and employing fixed molecules can reduce model accuracy. Adhering to MLR premises aids in validation. Analysis of secondary data from three articles showed that all five MLR models were invalid, emphasizing the need to verify MLR assumptions before utilizing the QSAR approach.