Artificial Intelligence-Driven Natural Product Discovery for Cancer Metastasis and Chemoresistance: From Computational Prediction to Preclinical Validation

TL;DR

AI tools are helping discover natural compounds that could treat aggressive cancers by targeting multiple biological processes.

Contribution

A framework for using AI to identify and validate natural products for cancer treatment, with a focus on metastasis and chemoresistance.

Findings

AI and machine learning can optimize and prioritize natural compounds for drug development.

Computational methods like graph neural networks and virtual screening improve ADMET and molecular analysis.

Case studies show AI-assisted discovery of natural products with in vivo and in vitro validation.

Abstract

Cancer is a major public health burden, with substantial impacts on survival and quality of life. It becomes particularly life-threatening when it spreads to other organs or develops resistance to treatment. Single-target therapies remain common in current cancer care, yet this strategy often fails to deliver long-term disease control. Natural compounds show promise as cancer therapies because they can simultaneously affect multiple biological targets. However, their complex structures, poor solubility, and less-than-ideal drug-like properties have slowed progress toward clinical approval. This review discusses how artificial intelligence and computational tools can help overcome these challenges by identifying, optimizing, and prioritizing natural compounds with anticancer potential. In this article, we have reviewed the recent computational methods, showcase real-world examples of successful discoveries, and offer a practical guide for choosing the best approach with the available data and resources. These advances could accelerate the development of more effective treatments for patients with aggressive or treatment-resistant cancers.

Cancer metastasis and chemoresistance are primary reasons for cancer-related mortality. Current therapeutic options rely mostly on single-target drugs, which often fail to exhibit long-lasting remission of the disease progression due to the complexity of metastasis and resistance mechanisms. Natural products (NPs) possess inherent structural diversity, rendering them suitable as multi-target agents. The utilization of NPs is often impeded in treating complex diseases such as cancer, even though approximately 65% of approved anticancer drugs are NP derivatives, or synthetic derivatives containing NP-pharmacophores, due to various factors, including poor aqueous solubility and variable oral bioavailability, structural complexity, synthetic inaccessibility, and stereochemical diversity that confounds structure–activity relationship analyses. This review discusses how integrating artificial intelligence (AI) and machine learning (ML) with chemoinformatics can identify, prioritize, and experimentally validate NPs, potentially paving the way for new drugs that address intricate processes such as metastasis and resistance. We summarize the recent computational advances in the field, including graph neural networks, attention mechanisms, Siamese networks, virtual screening, and network pharmacology. These advancements address ADMET optimization, molecular representation, virtual screening, network pharmacology, and experimental validation. We emphasize how each of these approaches tackles the unique challenges associated with NPs. We contextualize our review within the specific challenges presented by the chemical space of NPs. Additionally, we analyze real-world case studies of successful AI-assisted NP discovery and categorize the quality of evidence into three levels: Level A, which includes in vivo efficacy with mechanistic details; Level B, which consists of in vitro validation of mechanisms and phenotypes; and Level C, which represents computational hypotheses that are awaiting experimental verification. Additionally, we propose an operational framework for selecting suitable AI methodologies based on available data, target characterization, and validation resources. Finally, we emphasize the limitations and future directions in AI-facilitated NP discovery.

Figures