Page 12 - TECH MAGAZINE CSE

AN Technical Magazine
VESHAN Maharaja Agrasen Institute of Technology

FACULTY
CORNER
Dr. Moolchand Sharma, Assistant
Professor, CSE Department, MAIT, Delhi

Department of Computer Science &
Engineering

BIO-INSPIRED OPTIMIZATION TECHNIQUES FOR DRUG DISCOVERY

The recent practical application of artificial intelligence (AI) has been accelerated by the
availability of advanced computer hardware, such as GPUs, enabling faster parallel
processing. The pharmaceutical industry has seen a significant rise in machine learning

(ML) applications, particularly in deep learning (DL). This technological shift is driven by
the need for efficient screening in pharmaceutical development, where 5,000–10,000
chemical compounds are typically screened for each new drug. From this, around 250

compounds advance to preclinical testing, and only about five proceed to clinical testing.
The entire process, from drug discovery to marketing, typically spans 10 to 15 years.
Traditionally, drugs were discovered through methods like extracting active ingredients.
Chemically synthesised compounds from this process play crucial roles in treating
diseases. Modern drug discovery includes candidate identification, synthesis, screening,

and optimisation for properties like affinity and efficacy. Successful compounds proceed
to clinical trials. The current constraints of existing models in drug target interaction
encompass several factors. It includes dependence on limited labelled data, difficulties in

interpretability, struggles in capturing the intricacies of biological systems, and the
potential for overfitting, prompting the proposal of two novel mechanisms for predicting
drug-target binding affinity. In this context, two novel mechanisms have emerged to
enhance drug-target binding affinity prediction.

The first mechanism, DeepFNN-DTBA, leverages deep learning to improve prediction
accuracy by addressing limitations in traditional approaches. It incorporates advanced
network architectures to provide robust and reliable affinity estimations. The second

mechanism combines a bio-inspired magnetic particle Swarm Optimization Algorithm
(MPSOA) with a CNN-AttBiLSTM framework. This approach integrates convolutional neural
networks (CNNs) to extract high-level local features and an Attention-based Bidirectional
Long Short-Term Memory (AttBiLSTM) network to capture semantic relationships and
long-term dependencies.