Research Highlights (Jun 2001)

Known for work in computational biology especially protein folding, his pioneering use of an all-atom potential energy function and Cartesian coordinate energy minimization on an entire protein [1,6] made molecular dynamics simulations possible. This also led to the popular Jack-Levitt method for refining coordinates against X-ray data [19]. He elucidated the nature of protein motion and established protocols for realistic simulation in solution [45]. His study of aromatic interactions as electrostatic in nature[44], complemented his earlier work on the importance of electrostatics in enzyme catalysis [10]. He co-discovered the four protein-fold classes [12] and explained how these segments pack [15,27]. He was also the first to automate secondary structure identification [14]. His basic studies of protein components continue with his simulation of hydrophobic interaction [114].

He also pioneered simulation of protein unfolding in solution [59,62], emphasizing qualitative aspects and using film to show protein motion [30]. Directly approaching protein folding, he introduced simplified representations [9], showing they could be used in energy calculations [11] and complete enumeration of folds [60]. His early use of restraints and annealing in folding forms the basis for current methods of NMR structure determination [35]. Recently, he showed that the amino acid sequence can distinguish native-like folds at low resolution [70,80]. This method was one of the best at ab initio folding at the 1998 Critical Assessment of Structure Prediction (CASP) meeting [95,108]. He has recently developed new methods to predict folded structures and two of these (with Samudrala and Keasar, respectively) were in the best six entries at the 2000 CASP meeting.

Primarily focused on proteins, he has also contributed to the computational structural biology of DNA and RNA. Starting with his model for tRNA [2], which captured many features of the subsequently determined x-ray structure, he went on to correctly predict that DNA would have 10.5 not 10 base-pairs per turn in solution [20]. He also ran the first molecular dynamics simulations of DNA in vacuo [32] and in solution [84].

Turning his attention to sequence/structure analysis and bioinformatics, he has classified folds in genomic sequences [86] and compared results of sequence alignment with those of structure alignment [90]. He has developed methods to combine distant homology searches with automated modeling [61,94]. These results have been combined in the BioSpace database [106,110].