Pode a IA substituir 60% da I&D farmacêutica ao projetar e testar novos fármacos in silico usando química generativa e modelos preditivos de toxicidade ?

Os modelos de deep learning como o AlphaFold já revolucionaram o dobramento de proteínas. A IA generativa está agora a propor moléculas inovadoras com afinidades de ligação promissoras — levantando a questão de quando a IA poderá assumir completamente a descoberta de fármacos.

Background

As of 2024, AI-driven generative chemistry and predictive toxicity models have made significant strides in accelerating early-stage drug discovery, enabling rapid in silico design and screening of molecular candidates. Techniques such as multi-objective optimization with reinforcement learning (e.g., REINVENT or MolGen) and transformer-based models (e.g., AlphaFold2-informed docking) can propose novel structures with favorable binding affinities and reduced off-target risks. Deep learning models like AlphaFold have already revolutionized protein folding. However, no published source supports the claim that these tools can autonomously replace 60% of traditional pharmaceutical R&D—clinical trials, regulatory filings, and large-scale human trials remain human-led and data-intensive. Current industry practice emphasizes AI as a force multiplier in hit discovery and lead optimization rather than a wholesale replacement of R&D workflows.