سامانه همایش های دانشگاه آزاد اسلامی واحد نجف آباد

صفحه اصلی / اولین همایش بین المللی تحول خدمات سلامت: تلفیق مراقبت، درمان، بازتوانی و امیدبخشی

Systematic Review of Machine Learning and Deep Learning Approaches for Target Binding and Drug Permeability Prediction

نویسندگان :

Maryam Shahin¹ Alireza Zarei²

1- موسسه آموزش عالی آپادانا 2- دانشگاه بین المللی امام رضا علیه السلام

کلمات کلیدی :

Machine Learning،Deep Learning،Artificial Intelligence،Protein Binding،Pharmacokinetics،Cell Membrane Permeability

چکیده :

Abstract: Background: The pharmaceutical development pipeline is constrained by high attrition rates stemming from late-stage pharmacokinetic and toxicokinetic failures. Accurately predicting target binding affinity (TBA) and cell membrane permeability is critical for early-stage screening. Recently, Machine Learning (ML) and Deep Learning (DL) have emerged as robust, data-driven alternatives to traditional, cost-prohibitive empirical assays, though a comprehensive synthesis of recent algorithmic advancements remains sparse. Method: Adhering strictly to PRISMA 2020 guidelines, a systematic literature search was conducted across PubMed, Scopus, IEEE Xplore, and Web of Science for studies published between 2025 and 2026. From an initial pool of 600 records, 23 foundational studies were selected following rigorous screening. Methodological quality was evaluated using the MINORS criteria, and data regarding ML/DL architectures, datasets, and performance metrics were synthesized narratively. Finding: The field has experienced a paradigm shift from 1D sequential descriptors to 2D Graph Neural Networks (GNNs) and 3D structural embeddings. For TBA prediction, attention-based GNNs represent the state-of-the-art, offering high accuracy (R^{2}\gt 0.85) and isolating atomic-level interactions. Concurrently, Multitask Learning (MTL) has become the gold standard for permeability prediction (e.g., Blood-Brain Barrier and Caco-2), simultaneously computing interdependent variables such as lipophilicity and efflux pump status to improve predictive fidelity. Conclusion: The integration of ML and DL has transitioned pharmacological profiling from rigid molecular docking to highly dynamic predictive pipelines. To ensure ultimate clinical adoption and validate biochemical logic, future architectures must integrate Explainable AI (XAI) and prioritize physics-informed neural networks paired with high-resolution 3D datasets.