On Thursday, October 30th, more than 300 people from academia and industry gathered in the auditorium to participate in the Boltsgen Seminar hosted by the Abdul Latif Jameel Clinic for Machine Learning in Health (MIT Jameel Clinic). The event was headlined by Hannes Stärk, an MIT PhD student and lead author of BoltzGen, who announced it just a few days ago.
Built on Boltz-2, the open-source biomolecular structure prediction model for predicting protein binding affinities that made waves over the summer, BoltzGen (officially released on Sunday, October 26) is the first model of its kind to take a step further by producing novel protein binders ready to enter the drug discovery pipeline.
This is made possible by three key innovations. One is VoltzGen’s ability to integrate protein design and structure prediction to perform a variety of tasks while maintaining state-of-the-art performance. Constraints built into BoltzGen are then designed based on feedback from wetlab collaborators to ensure that the model creates functional proteins that do not violate the laws of physics or chemistry. Finally, a rigorous evaluation process tests models targeting “untreatable” diseases and pushes the limits of BoltzGen’s binder generation capabilities.
Most models used in industry and academia are capable of either structure prediction or protein design. Moreover, they are limited to producing certain types of proteins that bind well to simple “targets.” As long as the training data is similar to the target during binder design, the model often works, just as students answer test questions like homework. However, existing techniques are mostly evaluated on targets where binder-containing structures already exist, resulting in poor performance when used on more difficult targets.
“There have been models that have tried to address binder design, but the problem is that these models are modality-specific,” Stark points out. “A general model doesn’t just mean it can handle more tasks. Additionally, physics emulation is learned by example, resulting in a better model for each individual task. Also, a more general training scheme can provide more such examples that contain generalizable physical patterns.”
Boltsgen researchers went out of their way to test Boltsgen against 26 targets, ranging from treatment-related cases to cases explicitly selected due to their dissimilarity to training data.
This comprehensive validation process, conducted in eight wet labs across academia and industry, demonstrates the breadth of this model and its potential for breakthrough drug development.
Parabilis Medicines, one of the industry collaborators who tested Voltugene in a wet lab setting, praised Voltugene’s potential and said, “We believe that incorporating Voltugene into our existing Helicon peptide computational platform capabilities will accelerate our progress in delivering innovative medicines for major human diseases.”
The open source releases of Bolz 1, Volz 2, and now Bolzgen (previewed at the 7th Annual Molecular Machine Learning Conference on October 22) bring new opportunities and transparency to drug development, but also suggest that the biotech and pharmaceutical industries may need to reevaluate their products.
As BoltzGen becomes a hot topic on social media platform “The time lag in the performance of a chat AI system from private to open is [seven] “It looks like it will be even shorter in the protein space,” Grace wrote in the post. How can companies become Binder-as-a-Service? [recoup] Why invest when you can wait a few months for the free version? ”
For those in academia, Boltsgen represents the expansion and acceleration of scientific possibilities. “A question we often hear from students is, ‘Where can AI change the game of treatment?’” says Regina Barzilay, senior co-author and MIT professor who is AI faculty lead at the Jameel Clinic and an affiliate of the Computer Science and Artificial Intelligence Institute (CSAIL). “We cannot change the situation unless we identify undruggable targets and propose solutions,” she added. “The emphasis here is on unanswered questions, and that’s what sets Hannes’ work apart from others in the field.”
“Models like VoltzGen, released completely open source, enable broad community-wide efforts to accelerate drug design capabilities,” said senior co-author Tomi Jaakkola, Thomas Siebel Professor of Electrical Engineering and Computer Science at Jameel Clinic and CSAIL.
Looking to the future, Stark believes that the future of biomolecular design will be transformed by AI models. “We want to build tools that help us manipulate biology to solve diseases and perform tasks using molecular machines that we haven’t even imagined yet,” he says. “We want to provide these tools to help biologists imagine things they never thought possible before.”
