Computational Molecular Medicine is a class I took Junior year of college. The course assumed a basic understanding of probability and statistics and went on to cover topics such as hypothesis testing, parametric and nonparametric tests of independence, and enrichment analysis. Though, the focus of the course was on statistical learning and how learning algorithms, in particular classifiers, are applied to modern medicine and cancer classification and discovery. The bias-variance tradeoff, ROC curves, cross-validation, and a few basic discriminative and generative algorithms were introduced before deep learning and the basis of neural networks was briefly discussed. Unsurprisingly, all of this was in the context of computational biology and applications to medicine.
A more concise and elegant description of what went on over the semester is better summed up by the professer, Dr. Donald Geman:
Computational systems biology has emerged as the dominant framework for analyzing high-dimensional “omics” data such as DNA variants and RNA concentrations. One major purpose is to uncover the perturbations in the genome, transcriptome and gene networks associated with disease. Today, computational methods are largely based on statistical hypothesis testing, machine learning, and stochastic graphical models. Our featured application is precision oncology. We apply these methods to annotating genomes, predicting disease states, clinical outcomes and treatment effects, as well as modeling disease progression, primarily in cancer.
The motivation and a description of the project itself can be found in the attached report, and the abstract of the paper is copied here:
Cancer classification is widely regarded as the first and most important step in the development of cancer diagnosis and treatment. For example, tumors which exhibit strikingly similar phenotypes may share an insiginificant number biological markers and lead to drastically different clinical treatments. This paper presents an approach to feature recognition and cancer classification while making use of standard statistical procedures and machine learning algorithms. Results of preliminary statistical analysis are convincingly effective at predicting significant features within the data set and provide a route to feature discovery in the future. Statistical learning algorithms prove less effective at class prediction while providing an excellent example of the bias-variance tradeoff within machine learning and the care researchers must take to consider the makeup of a data set as it applies to training a classifier and mirroring reality.