From Idea to Action
Founded by a team of Mila students and researchers, Dreamfold is a pioneering protein design startup at the intersection of health research and innovation. The company's roots trace back to the turbulent days of COVID-19, when Maksym Korablyov, co-founder and CEO of Dreamfold, envisioned a project at Mila to explore drug design solutions for the pandemic. From there, the project sparked a new vision: harnessing the power of generative modeling to revolutionize protein design. This idea evolved into Dreamfold, a startup born from Mila’s collaborative environment and blending the talents of computational scientists, chemists, and biologists, in order to push the frontiers of protein modeling.
Leveraging World-Class Expertise
Dreamfold is on a mission to transform protein design through cutting-edge generative modeling. Their goal is to create models that excel not only computationally but also in the lab, delivering practical, biologically relevant results. By combining the diverse expertise of their team, Dreamfold aims to tackle complex biological challenges and achieve breakthroughs in protein design. A major achievement for the team has been the development of FoldFlow-2, an advanced generative model designed to enhance the accuracy and functionality of synthesized proteins.
“Because proteins are so rich in their structure and sequences, we needed a generative model that took advantage of both the 3D structure, as well as the huge database of protein sequences we have,” explained Joey Bose, Machine Learning Scientist at Dreamfold, “That was the inception for FoldFlow-2: leveraging protein sequences as well as 3D structures in order to create a multimodal generative model.”
As of summer 2024, FoldFlow-2 is currently the leading protein generative model, outperforming previous models like RoseTTAFold Diffusion from the University of Washington.
A Lab-Grown Success
In just a few short years, Dreamfold has achieved significant milestones. The team has published several groundbreaking papers, including spotlight presentations at major conferences like ICLR and ICML. Gaining recognition as the best protein generative model available today, FoldFlow-2 stands out for its ability to generate diverse, novel proteins unconditionally — in other words starting from random noise — and with high potential for lab success. Their models’ initial protein synthesis results are promising, laying a strong foundation for further biological validation and application in health research and pharmaceuticals. Dreamfold aspires to set new standards in the field, looking towards DeepMind's AlphaFold and its achievements in protein structure prediction as inspiration.
The Mila Advantage
When asked what gives Dreamfold its advantage, the team is unanimous. “It’s all about the people, says Joey Bose, “Really, the caliber of people that we have is top-notch. We have protein experts, we have generative modeling experts. And just the fact that we're at Mila makes it easy to access a lot of cool people.”
The Dreamfold team repeatedly echoes how instrumental Mila’s support has been to Dreamfold’s journey. Among Mila’s many advantages, the entrepreneurs cite their incredible proximity to diverse research expertise and the freedom to pursue ambitious projects without bureaucratic constraints, in addition to unparalleled advisory support from esteemed figures like Yoshua Bengio.
The institute’s Entrepreneurship Lab has also provided essential support over the startup’s development, from organizing international networking trips to facilitating potential collaborations and hosting workshops and events. And of course, Dreamfold’s entire team is composed of current and former Mila members, hailing from Mila’s rich, collaborative research community.
Jarrid Rector-Brooks, co-founder and CSO, jokes: “Our entire team is made up of Mila people — all the scientists, everybody's a Mila person!” And when the laughter dies down in the group, his tone is more serious, grateful: “There are just so many good people in machine learning and so many other fields at Mila. The support from Mila has really been great.”
