Revolutionizing Drug Discovery with Artificial Intelligence
Artificial intelligence (AI) is ushering in a new era in drug discovery, and institutions like Mount Sinai are at the forefront of this transformation. Dr. Ming-Ming Zhou, a renowned researcher in physiology and biophysics at the Icahn School of Medicine at Mount Sinai, reflects on how therapeutic research has evolved from structure-based approaches to AI-driven methodologies. His pioneering work in targeting bromodomains—a protein family involved in gene transcription—has laid the foundation for innovative cancer and inflammation therapies.
“Most breakthrough discoveries are made from existing evidence,” explained Zhou. “AI allows us to connect the dots in novel ways to solve problems.” This mindset is steering Mount Sinai’s ambitious efforts to democratize AI in pre-clinical drug development.
Mount Sinai’s AI Small Molecule Drug Discovery Center
Launched in April, the AI Small Molecule Drug Discovery Center is spearheaded by Dr. Avner Schlessinger, a professor of pharmacological sciences and associate director of the Mount Sinai Center for Therapeutics Discovery. The center aims to revolutionize the drug discovery process by leveraging computational power to identify, optimize, and repurpose drug candidates with unprecedented speed and accuracy.
Through seminars, internships, and hackathons, the center is committed to providing hands-on training to the next generation of scientists. By fostering collaborations with pharmaceutical giants, biotech firms, and academic institutions, Mount Sinai seeks to build a vibrant AI research ecosystem.
“Drug discovery is inherently inefficient,” Zhou emphasized. “We need new platforms that encourage collaboration and out-of-the-box thinking. This center is our answer to that challenge.”
Three Pillars of Innovation
The center’s research focuses on three main areas: designing novel compounds using generative AI, optimizing existing drugs for improved efficacy and safety, and predicting drug-target interactions to explore new applications for known molecules.
“I’ve been working in AI since before it was mainstream,” said Schlessinger. “Now, with access to Mount Sinai’s extensive datasets, we can create real solutions that impact patient care.”
Mount Sinai’s integration of AI into its research is particularly impactful because it operates within a hospital system. This enables researchers to work on highly translational projects, from Alzheimer’s disease target identification to mutation pathogenicity prediction using patient data.
Education and the Future of AI in Medicine
Dr. Marta Filizola, dean of Mount Sinai’s graduate school of biomedical sciences, leads the educational arm of the initiative. Under her leadership, Mount Sinai has launched a PhD program in Artificial Intelligence and Emerging Technologies in Medicine (AIET). “We’ve built an infrastructure to enhance AI visibility and provide students with hands-on opportunities to conduct research that directly improves human health,” she said.
Data Accessibility: A Critical Barrier
While public resources like the Protein Data Bank (PDB) have supported AI advancements such as AlphaFold, many potential drug targets remain outside of these databases. This gap has led AI-focused biotech firms to generate their own proprietary datasets, which often remain inaccessible to the broader research community.
“Benchmarking AI models without access to proprietary data is a major challenge,” noted Robin Roehm, CEO of Apheris. His company is working with the AI Structural Biology Consortium and pharmaceutical leaders like AbbVie and Johnson & Johnson to fine-tune OpenFold3, a protein structure prediction algorithm, in a secure environment using confidential data.
Open-Source Efforts: The Case of Boltz-2
In contrast to proprietary models, researchers at MIT’s Jameel Clinic have released Boltz-2, a freely available tool that predicts molecular binding affinity with both speed and accuracy. Developed in collaboration with Recursion, Boltz-2 is offered under the permissive MIT license, enabling commercial and academic use alike.
This move was, in part, a response to the limited accessibility of AlphaFold 3, whose code was initially not released, prompting a significant backlash from the scientific community. More than 1,000 scientists signed a protest letter demanding transparency, and the developers eventually released the code under a non-commercial license.
“Just when it seemed like closed models would dominate, the community came together around Boltz-2,” said MIT’s Corso. “It was a powerful moment for open science.”
Performance and Impact of Boltz-2
Boltz-2 excels in both speed and precision. It was the top performer in binding affinity prediction at the CASP16 competition in December 2024, achieving calculations in just 20 seconds—over 1,000 times faster than traditional methods. This capability is crucial, as binding affinity dictates a drug candidate’s progress through the development pipeline.
Najat Khan, chief R&D and commercial officer at Recursion, praised Boltz-2’s release: “This is a significant step forward in integrating technology with biology and chemistry. Binding affinity is central to drug development, and Boltz-2 addresses this challenge head-on.”
The Road Ahead
Despite the ongoing issues surrounding proprietary data and commercial interests, the combined efforts in education, collaboration, and open-source modeling are fostering a more inclusive scientific landscape. Mount Sinai’s endeavors exemplify a commitment to not only technological innovation but also democratizing access to life-saving research tools.
Whether AI-driven drug discovery will ultimately become a truly democratic process remains to be seen. However, initiatives like Boltz-2 and Mount Sinai’s new center are certainly pushing the field in that direction.
This article is inspired by content from Original Source. It has been rephrased for originality. Images are credited to the original source.