Proteins
From fitness prediction and uncertainty to sequence and structure generation with experimental feedback.
Quine is a Microsoft Research program building AI systems that help scientists reason across biological evidence, choose experiments, and learn from what happens in the lab.
The Quine overview video will be added here.
Biological research often advances through slow cycles of proposing, testing, and revising ideas. A single experiment can take weeks or months, and negative results still consume scarce samples, time, and attention. We are exploring whether better computational guidance can help researchers consider more possibilities and use physical experiments only where they provide the most useful information.
The model is being developed to connect evidence across proteins, cells, tissues, and genomes. The tool layer is being developed to help researchers break questions into steps, use models and scientific tools, compare evidence, and revise a plan. Microsoft scientists, collaborators, and experimental partners evaluate the system through real-world biological research questions.
These parts are developed together. Scientific questions reveal which capabilities and data matter. Model and tool improvements expand what researchers can test. Experimental results provide evidence about what worked, what failed, and what researchers should improve or explore next.
Quine develops through use. Research programs test whether it helps scientists reason across evidence, choose informative experiments, and respond to unexpected results. Those results show where the system is useful, where it falls short, and what the team should build next.
Quine builds on years of research at Microsoft across proteins, regulatory genomics, cancer cell state, pathology, microscopy, and biological data. Our research journey follows how those independent lines of inquiry developed, challenged assumptions, and began to inform a shared system.
From fitness prediction and uncertainty to sequence and structure generation with experimental feedback.
Using evolutionary signals to study functional constraints and computationally generate and evaluate regulatory sequences.
Connecting cell state, environmental context, drug response, and scalable experimental systems.
Learning from whole-slide images, multiple magnifications, clinical-domain text, and genomic biomarker information.
Distinguishing biological signal from batch effects, neighboring-cell context, and other misleading patterns in experimental data.
Studying how dataset composition, comparison, and scale shape what models learn and how well they generalize.
The research journey traces six lines of work across biology and machine learning, showing where they overlapped and how we arrived at Quine.
Contact the team about Quine, our research, or the stories published here.
Please do not submit confidential, personal, clinical, or proprietary research information by email.