Digital tools and predictive models gain ground in drug discovery

Digital drug design tools and AI are being integrated into drug discovery as typical timelines for successful drugs stretch from 10 to 14 years, or sometimes even longer, to bring a drug from discovery to approval. The cost of bringing a drug to market can range from hundreds of millions to multiple billions of dollars, and only about 12% of drugs that reach clinical trials will gain FDA approval.

Published estimates of R&D cost per approved drug vary widely depending on methodology and what is included. One influential analysis estimated costs of approximately $1.4 billion out-of-pocket and $2.6 billion capitalized in 2013 dollars, while a public-data analysis of FDA approvals estimated the median capitalized research and development investment to bring a new drug to market at $985.3 million and the mean investment at $1,335.9 million in the base case analysis. Financial costs for a loss associated with a failed late-stage trial, including sunk development costs, can range from $800 million to $1.4 billion.

Clinical development success rates vary by disease and dataset. A large benchmark of programs from 2011–2020 estimated an approximate 7.9% likelihood of approval from Phase I to approval overall, with oncology lower at approximately 5.3%. Only a small fraction of compounds entering preclinical testing progress to first-in-human studies.

The long timelines and high costs associated with drug discovery have led to a gap in medical needs for patients with rare diseases. Although 3.5% to 5.9% of the world’s population, about 263 million to 446 million people, suffer from rare diseases, 95% of them do not have FDA-approved treatment options in the U.S. With such high failure rates leading to wasted money and time, most trials are focused on more common diseases that promise a bigger payoff if the drug makes it to market.

Today, digital drug design encompasses molecular simulations that model how compounds interact with biological targets, machine learning models trained on vast chemical libraries to predict drug candidates, generative AI that proposes entirely novel molecular structures, and lab automation platforms that connect computational predictions to physical experiments. In the 2020s, AI and machine learning were integrated into computer-aided drug design approaches, and deep learning models enabled unprecedented predictive accuracy. AlphaFold and other generative AI tools transformed structure prediction and molecular design.

Before the 1970s, drug discovery was dominated by serendipitous observations and trial-and-error screening. Beginning in the 1970s, computer-aided drug design using computational techniques emerged, and applications gradually increased throughout the 1980s and beyond. In the 2000s, structural biology, genomics and bioinformatics began to drive computer-aided drug design development, protein and chemical libraries expanded exponentially, and virtual screening became standard practice.

The first AI-designed drugs are currently advancing through clinical trials, and companies are demonstrating end-to-end AI pipelines. One estimate projects the AI drug discovery market to grow from $4.6 billion in 2025 to $49.5 billion by 2034 at a 30% compound annual growth rate.

A recent paper analyzed information-theoretic predictive models on a set of biological sample screening results for drug design. Unlike existing tools that primarily provide probabilistic biological activity spectra from SMILES input, the proposed system incorporates a hybrid neural network integrated with information-theoretic predictive models to dynamically simulate pharmacokinetic and pharmacodynamic processes using structure-based and biological sample data. Based on the set of features of biological samples, two active samples were selected, which, with a high degree of reliability, are also suitable for creating drugs based on the formyl peptide receptor.

The paper said researchers can choose the most important characteristics of biosamples with the use of class information portraits, which lowers the effort required for drug development and improves forecast accuracy. Feature contribution analysis identified features that positively affect the active state of bio samples, and the strength of their influence was also determined.