<p>Drug discovery has been one of AI's most hyped application areas for half a decade. The promise: AI can search molecular space vastly more efficiently than traditional high-throughput screening, identifying promising candidates in months rather than years. The problem: promising candidates have been abundant; candidates that survive clinical trials have not. That is beginning to change.</p>
<h2>The Three Phase II Candidates</h2>
<p>Insilico Medicine's ISM001-055, targeting idiopathic pulmonary fibrosis, was the first AI-designed drug to reach Phase II and has now reported positive interim data — patients on the compound show statistically significant improvement in lung function decline rate compared to placebo. The compound was identified and optimized in 18 months using Insilico's generative chemistry platform, versus the typical 4-6 years for conventional discovery.</p>
<p>Recursion Pharmaceuticals has two candidates in Phase II: REC-994 for cerebral cavernous malformation and REC-2282 for NF2-mutated meningioma. Both emerged from Recursion's phenomics platform, which uses cellular imaging at massive scale combined with machine learning to identify disease mechanisms and candidate compounds simultaneously.</p>
<h2>What Phase II Means</h2>
<p>Phase II trials test efficacy in a larger patient population (typically 100-300 patients) after Phase I confirms safety. The failure rate at Phase II is still high — approximately 65% of compounds that enter Phase II fail to advance. But reaching Phase II with AI-discovered compounds at all validates the earlier pipeline steps (target identification, hit generation, lead optimization) in a way that preclinical data could not.</p>
<h2>The Timeline Question</h2>
<p>Even optimistic scenarios see the first AI-designed drug reaching market no earlier than 2027-2028. But the cost implications are significant: AI-assisted discovery has reduced pre-Phase I investment for these compounds to approximately $10-15M versus the industry average of $50-100M. If Phase II and III success rates hold comparable to conventionally discovered drugs, the total development cost reduction could be 40-60%.</p>