With the extensive development of genomic sequencing and rapid protein expression, there are now many more targets available for protein structure determination than can be completed by even the most advanced automated methods. Many of these targets also present well-known experimental difficulties for the traditional methods of crystallography and NMR. However, help in interpreting the genomic data is available from computational techniques that use knowledge extracted from the database of experimentally determined structures, in order to predict structures for proteins with different sequences. The effective use of these computational techniques would be greatly enhanced by the ability to rapidly discriminate among and confirm predictions with experimental data that can be acquired rapidly and easily. For example, appropriate experimental methods can provide measurements of distance (e.g. by chemical cross-linking of specific residues), assays of residue accessibility (e.g. by chemical modification of surface residues), and low-resolution characterizations of shape (e.g. by solution scattering).

We are developing a comprehensive computational-experimental protocol for the rapid discrimination of predicted protein structure via multiple modes of experimental data. The multimodal discrimination approach poses challenging computational problems in combining rather sparse, noisy information from different experimental methods and describing quantitatively the extent of discrimination provided. Associated reasoning algorithms overcome problems of noise and sparsity by uncovering consistent sets of features, targeting clarifying queries in response to conflicts, and planning additional experiments.

• Aly Azeem Khan
• Shobha Potluri
• Chetak Sirsat
• Tim Williamson
• Xiaoduan Ye

• Alan Friedman (biology, Purdue)
• Janusz Bujnicki (IIMCB bioinformatics)