Abstract

Suggesting target proteins for compounds that give rise to a particular cellular phenotype but have unknown protein targets is crucial in drug research. This can be done through experimental target finding methods, or via computational approaches. Computational methods are increasingly gaining preference because they are less time consuming and reduce hypothesis space to a smaller number of testable biological targets. Ligand-based approaches, which form a subgroup of target prediction algorithms, mine large bioactivity databases and employs pattern recognition/machine learning techniques to find the target protein associated with the compound. To this end, I implemented two in silico protocols and compared their performance. The first protocol compared two probabilistic models by a variety of performance measurements, being the Naïve Bayes classifier and the inverse Ising model. The second protocol addresses the promiscuity of the bioactive compounds.