Prognosis AI represents a breakthrough approach to disease prediction and therapy selection using transcriptomic data. Built on Biostate’s Biobase foundation model—which has been pre-trained on massive unlabeled RNA datasets—Prognosis AI applies the same revolutionary approach that transformed AI language models to biological data. Just as large language models like GPT learn the patterns and structure of human language before fine-tuning on specific tasks, Biobase learns the fundamental ‘grammar’ of biology from raw transcriptomic data, enabling Prognosis AI to make powerful predictions with remarkably small sets of labeled clinical samples
Biobase is our proprietary foundation model trained on hundreds of thousands of unlabeled transcriptomic samples. Similar to how ChatGPT learns language patterns from massive text data, Biobase learns the underlying patterns and relationships in RNA expression data across diverse tissues, conditions, and species. This pre-training allows the model to develop a deep understanding of biological states before ever seeing disease-specific data.
Prognosis AI fine-tunes the Biobase foundation model using de-identified samples with clinical annotations. This two-stage approach—pre-training on unlabeled data followed by fine-tuning on smaller labeled datasets—allows us to create highly accurate predictive models with far fewer labeled samples than traditional approaches would require. This is particularly valuable for rare diseases or situations where labeled samples are difficult to obtain.
Our internal studies have demonstrated remarkable results. In hepatotoxicity prediction in rodents, our approach improved accuracy from 65% with traditional machine learning to 89% with our Biobase-Prognosis AI approach. For acute myeloid leukemia (AML), we’ve developed models that can predict which patients will respond to specific therapies with significantly greater accuracy than current clinical methods, potentially sparing patients from ineffective treatments while identifying beneficial options that might otherwise be overlooked.
If your organization has RNA samples with clinical annotations—whether for cancer, autoimmune conditions, transplant outcomes, or other disease areas—we can collaborate to develop custom Prognosis AI models. Our approach requires significantly fewer labeled samples than traditional methods, making it feasible to create predictive models even for rare conditions or small patient cohorts. We handle all aspects of model development while maintaining strict data privacy and security.
Your biobank of samples represents untapped potential for developing groundbreaking predictive models. Whether you have hundreds of samples from a specific disease cohort or a smaller collection of rare disease samples, our Biobase-Prognosis AI approach can likely generate valuable insights with your existing data. Contact us to discuss how we can collaborate to unlock the predictive power of your samples.