LOCATION: La Jolla, California
SUMMARY: The Cancer Cell Map Initiative (CCMI) is building an artificial intelligence for cancer. Using state-of-the-art machine learning approaches, this system will not only be able to predict clinical and therapeutic outcomes but will also explain the underlying biological mechanisms and thereby guide future experimental studies.
PROBLEM SPACE: A key challenge in cancer research is how to translate information about a patient into a correct diagnosis, appropriate treatment and ultimately a cure. Increasing amounts of data can now be collected, including complete tumor genomes, molecular profiles of RNAs and proteins, and extensive clinical histories including laboratory measurements and sophisticated scans. In what has been a surprise to many scientists and clinicians, analysis of tumor genomes has found that most tumor mutations are quite rare; hardly ever are the the same mutations present from one patient to the next except for a few that fall within well-known cancer genes. The key word used for this situation is heterogeneity: every tumor is, in a real sense, unique. In addition, sequencing of individual tumor cells has shown that such heterogeneity exists not only from one patient to another but also from one tumor cell to another even within the same tumor. Even then at the level of a single cell, the causes of cancer are not simple, attributable to only one mutation or gene. The causes are complex and multifactorial, involving systems of genes and environmental forces that must be regulated or dysregulated at multiple points to initiate or promote cancer.
SOLUTION: "Recent studies have found that biological networks can help to overcome the problem of genetic heterogeneity. Rather than such-and-such nucleotide is mutated, we discover this membrane complex is mutated or that pathway is mutated or even this cell type is mutated. In this way, networks recognize that patients with different, seemingly rare mutations are in fact mutated in common transcriptional networks, protein complexes or pathways.
To take advantage of these networks, the CCMI will first generate a comprehensive maps of the key protein-protein and genetic interactions underlying cancer. Protein-protein interaction maps the complete set of proteins that bind to another protein tell us about the physical structure of cancer cells. Genetic interaction maps knowing how deleting one gene impacts how cells respond to the loss of another gene tell us about how groups of genes function as pathways and networks. These pairwise molecular maps will be used to build hierarchical models, providing key insights into the cell's inner workings. Perhaps more importantly, this hierarchical model can be used to build a constrained neural network that will not only be able to predict biological outcomes but will also provide insights into the underlying biological mechanism driving outcomes."
CONTACT: [email protected]