Integrative Genomics as a Case Study for Integrative AI

As biology becomes an information science and information science becomes dominated by machine learning and other AI methods, it stands to reason that biology is becoming dominated by AI. To grapple with the systemic nature of disease and aging, it is necessary to do simulation modeling, data analysis, and machine reasoning regarding the multiple body subsystems across numerous datasets.

This emerging paradigm of medicine has been termed “P4 medicine that is predictive, preventive, personalized, and participatory” by systems biology godfather Leroy Hood. Osiris CEO and cofounder Dr. Ben Goertzel was an early practitioner of this view; since 2000 he has applied machine learning and other AI technologies to longevity and genomics, including in collaborative work with the CDC, NIH, and various universities.

In this spirit, the Osiris AI team has chosen biomedical data analytics—in particular the analysis of genomics data regarding longevity and age-associated diseases—as an initial testing ground for integrating multiple AI paradigms within the IBM Watson framework.

MOSES is used to find patterns in genomic datasets. The small programs representing these patterns are then imported into the AtomSpace hypergraph representation. Next, the PLN logic engine is used to draw conclusions by combining the patterns with knowledge obtained from biological ontologies like the Gene Ontology project, MSigDB, etc. and with knowledge extracted from biological texts using IBM Watson natural language processing technology.

For example, when applied to genomic data obtained from exceptionally long-lived people, MOSES can tell us what genes or what combinations of genes tend to have the most significant influence on these peoples’ long lives. PLN and reasoning about these MOSES models, together with other knowledge, can give us the hypotheses about how these genes impact aging. This can be a powerful tool for suggesting new experiments to run and for suggesting diagnostics to identify a disease state or predict future disease or longevity. It can also be applied to discover targets for either conventional drug therapy or gene therapies such as CRISPR.

In 2019, the Osiris bio-AI team will release a series of publications describing novel discoveries about aging and disease that have been uncovered using these methods during its 2018 research. However, these exercises in AI refinement and prototyping have importance going beyond these particular results and this particular subdomain. These methods will serve as part of the AI core of the Osiris Healthtech Studio project, and they also have general applicability beyond health-tech.

For instance, in financial services, there is a demonstrated value to applying the MOSES learning engine to combine price, volume, global macro, company accounting, and news sentiment data into combinational predictive models. Financial text analysis software is relatively mature, and an extensive amount of structured data pertaining to listed companies and their internal structures and external involvements is available. The methodology refined by the Osiris research team in the context of genomics AI will be adapted to play a crucial role in the Osiris Studio fintech module