Schizophrenia: A Molecule Designed with ChatGPT, No Trials

Key Takeaways

1Douglas Yao, a doctor in computational biology, claims to have developed PAC-3310 with ChatGPT in his garage

2The only available data concerns tests on rodents, with no scientific publication or clinical trial

3The development of a drug requires much more extensive evidence and validations, as illustrated by the case of Cobenfy

Why it matters : Announcements about new AI-derived treatments must be subjected to the requirements of scientific and regulatory validation before any human application.

A researcher trained at UCLA and Harvard claims to have designed a compound called PAC-3310 for schizophrenia in his garage, with the help of ChatGPT. At this stage, there are no peer-reviewed publications or clinical trials to support these claims, and the lengthy regulatory pathway for a drug remains intact.

Limited Evidence Against a Heavy Regulatory Burden

No data on PAC-3310 has been published in a peer-reviewed journal, and the compound has not undergone any clinical trials. The available information comes from experiments conducted on rodents and verified only by Douglas Yao himself. He highlights a reduction in MK-801-induced hyperlocomotion in mice, a marker that does not demonstrate efficacy in humans. By comparison, Cobenfy has gone through three phases of clinical trials in Bristol Myers Squibb's EMERGENT program, involving thousands of patients. The Tufts Center for the Study of Drug Development estimates that bringing a molecule through the regulatory process costs an average of $2.6 billion and takes 10 to 15 years, with a success rate of 12%. Even if presented to the FDA, a dossier based on tests conducted solo in a garage would not meet the usual validation requirements.

What Douglas Yao Claims to Have Done and Shown

On September 8, Douglas Yao posted a viral message on X that reached 6 million views, showcasing a bottle of yellow powder called PAC-3310. He claims to have designed this molecule with the help of ChatGPT and synthesized it in an improvised lab set up in his garage. At this stage, the trials have involved rodents, and Pace Pharmaceuticals has only one employee—himself. According to him, PAC-3310 caused no side effects in mice, even at a dosage ten times higher than the therapeutic dose. He also argues that AI could reduce the cost of drug development by a factor of a thousand.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Clinical Context: Cobenfy Approved, Distinct Mechanism, and Side Effects

Douglas Yao's target is Cobenfy, an antipsychotic combining xanomeline and trospium chloride, approved by the FDA for schizophrenia in September 2024. This treatment is presented as the first antipsychotic not to act by blocking dopamine since the 1950s, a period during which treatments relied on this mechanism following the discovery of chlorpromazine. However, Cobenfy is associated with notable side effects, including gastrointestinal—nausea, vomiting, constipation—and cardiovascular—hypertension, tachycardia.

Researcher's Background and Resources, Limitations of an Amateur Lab

Douglas Yao holds a PhD in computational biology and has experience in drug design. He has worked at UCLA and Harvard under the guidance of Alexander Gusev and Alkes Price. His publications in bioinformatics are well-cited, and his GitHub features projects in computational genetics. He claims to have taught himself synthetic chemistry. His home lab does not meet any pharmaceutical biosafety standards, although it is equipped with a $4,000 OpenTrons OT-2 robot for liquid handling. He further claims to have designed several thousand new therapeutic molecules last year and is also developing a new drug for Alzheimer’s.

AI Already Present in Molecular Prediction

Several artificial intelligence tools are used to predict molecular structures necessary for discovery work, such as Google’s AlphaFold, Meta’s ESMFold, or MIT’s DiffDock. However, this step does not equate to obtaining an approved drug.