7.91 Amy Keating Methods for Protein structure prediction Homology modeling Fold recognition Next time: Ab initio Prediction
Methods for Protein Structure Prediction Homology Modeling & Fold Recognition Next time: Ab Initio Prediction 7.91 Amy Keating
Review -Homology modeling Identify a protein with similar sequence for which a structure has been solved (the template) Align the target sequence with the template Use the alignment to build an approximate structure for the target Fill in any missing pieces Fine-tune the structure Evaluate success An excellent review Marti-Renom et al. Annu. Rev. Biophys. Biomol. Struct. 29 (2000): 291-325
Review - Homology Modeling • Identify a protein with similar sequence for which a structure has been solved (the template) • Align the target sequence with the template • Use the alignment to build an approximate structure for the target • Fill in any missing pieces • Fine-tune the structure • Evaluate success An excellent review: Marti-Renom et al. Annu. Rev. Biophys. Biomol. Struct. 29 (2000): 291-325
EDN CRABPI NM23 100 these numbers are from an mu>uHm-08 entirely O TEMPLATE· TARGET automated 50 ● MODEL· TARGET process- can do better with TEMPLATE- TARGET DIFFERENCE manua ALIGNMENT ERROR intervention 020406080100 SEQUENCE IDENTITY Marti-Renom et al. Annu. Rev. Biophys. Biomol. Struct. 29 (2000) 291-325 Courtesy of Annual Reviews Nonprofit Publisher of the Annual Review of TM Series. Used with permission
these numbers are from an entirely automated process - can do better with manual intervention Marti-Renom et al. Annu. Rev. Biophys. Biomol. Struct. 29 (2000): 291-325. Courtesy of Annual Reviews Nonprofit Publisher of the Annual Review of TM Series. Used with permission
Homology Modeling on a Genomic Scale · Requires automation Can't choose templates or fine-tune the alignment by hand MODBASE and 3D-CRUNCH http://alto.compbio.ucsfedu/modbase-cgi/index.cgi http://www.expasy.ch/swissmod/sm_3dcrunch.htm Automatic assessment is critical how reliable is the model?
Homology Modeling on a Genomic Scale • Requires automation – Can’t choose templates or fine-tune the alignment by hand! • MODBASE and 3D-CRUNCH http://alto.compbio.ucsf.edu/modbase-cgi/index.cgi http://www.expasy.ch/swissmod/SM_3DCrunch.html • Automatic assessment is critical - how reliable is the model?
One approach to assessment Want to compute the probability that a prediction is good based on properties of the model For a given score of the model e.g. Q-score-more on this later), use a training set of known examples together with Bayes'rule P(AB) =P(AA B/P(B)=P(AP(BA)RP(AP(BA)+ P(AP(B !A)y Assume probability of a good Vs a bad model is the same . e P(a)=P(A) where a good model; !a =bad model; B=Q-score P(good Q-score)=P(Q-scorel good)/P(Q-score good)+P(Q-score bad)] good models bad models Q -score Sanchez, R, and A sali. "Large-scale Protein Structure Modeling of The Saccharomyces Cerevisiae Genome Proc Nat/ Acad SciU SA. 95, no. 23(10 November 1998 ) 13597-602
One approach to assessment Want to compute the probability that a prediction is good, based on properties of the model For a given score of the model (e.g. Q-score - more on this later), use a training set of known examples, together with Bayes’ rule P(A|B) = P(A ^ B)/P(B) = P(A)P(B|A)/{P(A)P(B|A) + P(!A)P(B|!A)} Assume probability of a good vs. a bad model is the same, i.e. P(A) = P(!A) where A = good model; !A = bad model; B = Q-score P(good|Q-score) = P(Q-score|good)/{P(Q-score|good) + P(Q-score|bad)} Prob. Q-score good models bad models Sanchez, R, and A Sali. "Large-scale Protein Structure Modeling of The Saccharomyces Cerevisiae Genome." Proc Natl Acad Sci U S A. 95, no. 23 (10 November 1998): 13597-602