In silico hypothesis discovery

As has been introduced in the preceding chapters, the biomedical and life sciences domains have experienced explosive growth in terms of the volume and complexity of data sets available for investigation. However, while such data has become more and more available to researchers, the state-of-the-art in terms of effectively using such resources for hypothesis formulation and testing remains extremely basic. Most hypotheses that are evaluated in the modern scientific setting are a function of the intuition or belief systems of an individual or team of investigators, often informed by years of experience and prior research. While these approaches have served the community well and led to many important discoveries, it is important to ask if there are alternative approaches to hypothesis discovery. In this chapter, we will briefly introduce the core concepts that underlie such an alternative, which is known as in silico hypothesis discovery. These methods employ domain-specific conceptual knowledge collections, such as ontologies or knowledge that can be extracted from the domain literature using machine learning or natural language processing methods, in order to reason upon and generate hypothesis corresponding to a data set in an extremely high throughput manner. While in silico hypothesis discovery methods remain very early in their development at the present time, they also hold great promise in terms of accelerating the pace, breadth, and depth of scientific discovery in the “big data” era.