DESIGN NEW POTENT BETULIN DERIVATIVES AS ANTICANCER AGENT USING MULTIPLE LINEAR REGRESSION AND ARTIFICIAL NEURAL NETWORK METHODS
Uripto Trisno Santoso1,*, Muhammad Khafi2, Samsul Hadi2, Bayu Bakti Angga Santoso3

1Department of Chemistry, Faculty of Mathematics and Natural Sciences, University of Lambung Mangkurat, Jl. A. Yani Km 36 Banjarbaru, South Kalimantan INDONESIA

2Department of Pharmacy, Faculty of Mathematics and Natural Sciences, University of Lambung Mangkurat, Jl. A. Yani Km 36 Banjarbaru, South Kalimantan INDONESIA

3Master Program in Pharmaceutical Sciences, Faculty of Pharmacy, Gadjah Mada University, Jl. Sekip Utara, Sleman, Yogyakarta 55281, INDONESIA

*Corresponding author, tel/HP: +6285868736288, email: utsantoso[at]ulm.ac.id

Abstract

Betulin, betulinic acid, and its natural and synthetic derivatives act specifically on cancer cells with low cytotoxicity towards normal cells. The purpose of this study is to establish a multiple linear regression (MLR) model for design new potent betulin derivatives indicating high activity against HT29 human colon cancer cells, and the artificial neural network (ANN) method was used for model refinement. Topological descriptors properties were calculated using Mordred Descriptor web. Microsoft Excel for Microsoft 365 (Version 2206) integrated with XLstat version 24.3.1 (Addinsoft) and equipped with XLSSTAT R() function was used to MLR and ANN modelling. The ANN models were developed using a RProp+ algorithm with the number of input layers were varied from 2 to 7 and hidden layers 2 to 10. The final MLR model yield a fit to the training set data (R2 = 0.910, R2adj = 0.877) and shown to perform well in internal and external validations.

Y = -5.043-0.257(SlogP)-3.62x10E11(SM1D)-0.039(ATSC1i)
+18.808(Xc6dv)+0.127(ATSC4p)+1.205(GATS7c)

Interestingly, model refinement using an ANN applying the same descriptors generated better models, but it has lowered the predictive power. Furthermore, the MLR and ANN models have a significantly different in prediction the structures of the new potent betulin derivatives.

Keywords: Multiple Linear Regression, Artificial Neural Network, Betulin, Cancer, Descriptor, Quantitative Structure-Activity Relationship (QSAR).

Topic: Physical & Theoretical Chemistry