SEA technology is motivated by two ideas: that proteins can be related by their pharmacology, and that one can exploit these network relationships to discover new targets for small molecules.
These powerful ideas led SEA to be featured among Wired magazine's top scientific breakthroughs of 2009.
SEA relates proteins by their pharmacology by first aggregating many tiny chemical similarity signals among the ligands. By then leveraging extreme value statistics, SEA both filters out the uninteresting signals, and normalizes the aggregate results against a reference background to predict the significance of the pharmacological similarity. Highly significant target relationships, that were unknown in the reference data, are a fertile ground for budding drug discovery. For the genesis of the Similarity Ensemble Approach (SEA), see Keiser et al., Nature Biotechnology, 2007 (» view).
After efficacy, adverse drug reactions are the second most common reason that drugs fail in costly clinical trials. Utilizing SEA we've developed a fully-automated method to screen for side effects among drugs, on a broad scale. Briefly, from a matrix encompassing 656 drugs versus 73 side-effect targets, 1042 SEA predictions were ultimately tested, resulting in 499 drug-to-side-effect-target predictions unknown to the method. Furthermore, the 151 new drug-target associations experimentally confirmed at Novartis resulted in 247 significant links to actual adverse drug reactions of patients. The cost reduction and effort focusing potential of this computational method for early prediction of toxic liabilities is enormous. See Lounkine & Keiser et al., Nature, 2012 (» view).
Through what target is a small molecule acting to acheive an effect? This simple question can consume massive amounts of resources strewn among an intricate web of dead ends and false hopes. Of 681 neuroactive compounds that modulated zebrafish response to light, the Similarity Ensemble Approach (SEA) was ultimately able to predict at least one target for 586 of them. To test the viability of a sample of the previously unknown SEA predictions, 20 target predictions were tested experimentally, and over half (11) had measurable activities ranging from 1 to 10000 nM. See Lagner et al., Nature Chemical Biology, 2011 (» view)
The traditional view that drugs specifically bind a single target is radically shifting as we discover drug promiscuity may be the norm. To predict new targets using SEA, we queried 3,665 investigational and FDA-approved drugs against a reference of 246 MDL Drug Data Report (MDDR) targets. Among 3,832 non-trivial SEA target predictions, we selected 30 promising candidates for experimental testing. 77% (23 of 30) of our predictions were experimental confirmed with Ki binding affinities between 1 and 15000 nM. Moreover, the new targets are sometimes illuminating for drug efficacy. For Doralese, the 18 nM affinity for the new target dopamine D4 was over 10 fold more potent than the affinity for the canonical targets of 611 and 226 nM for α1A and α1B, respectively. For more insights into drug action gleaned by SEA, see Keiser & Setola et al, Nature, 2009 (» view). Here at SeaChange, we are actively expanding the core SEA technology to shortcut the discovery of new therapeutic uses for known drugs.
Classic pharmacology inferred the existence and distinction between drug targets by the relationships of ligands acting in different tissues. Modern chemoinformatics methods such as SEA are now returning to those ideas, effectively harvesting ligand relations to map out the poly-pharmacological landscape between small-molecule ligands and therapeutic targets. The resulting pharmalogical networks offer many low-hanging target associations that are both new and relevant, waiting to be discovered and tested. For more, see Keiser, Irwin, Shoichet, Biochemistry, 2010 (» view).