Transcript Slide 1
Structure Prediction Tertiary protein structure: protein folding Three main approaches: [1] experimental determination (X-ray crystallography, NMR) [2] Comparative modeling (based on homology) [3] Ab initio (de novo) prediction (Dr. Ingo Ruczinski at JHSPH) Experimental approaches to protein structure [1] X-ray crystallography -- Used to determine 80% of structures -- Requires high protein concentration -- Requires crystals -- Able to trace amino acid side chains -- Earliest structure solved was myoglobin [2] NMR -- Magnetic field applied to proteins in solution -- Largest structures: 350 amino acids (40 kD) -- Does not require crystallization Steps in obtaining a protein structure Target selection Obtain, characterize protein Determine, refine, model the structure Deposit in database X-ray crystallography http://en.wikipedia.org/wiki/X-ray_diffraction Sperm Whale Myoglobin PDB New PDB structures • April 08, 2008 – 50,000 proteins, 25 new experimentally determined structures each day Old folds New folds Example 1wey Ab initio protein prediction • Starts with an attempt to derive secondary structure from the amino acid sequence – Predicting the likelihood that a subsequence will fold into an alphahelix, beta-sheet, or coil, using physicochemical parameters or HMMs and ANNs – Able to accurately predict 3/4 of all local structures Structure Characteristics Beta Sheets Ab Inito Prediction Secondary structure prediction Chou and Fasman (1974) developed an algorithm based on the frequencies of amino acids found in a helices, b-sheets, and turns. Proline: occurs at turns, but not in a helices. GOR (Garnier, Osguthorpe, Robson): related algorithm Modern algorithms: use multiple sequence alignments and achieve higher success rate (about 70-75%) Page 279-280 Table Frequency Domain Neural Networks Training the Network • Use PDB entries with validated secondary structures • Measures of accuracy – Q3 Score percentage of protein correctly predicted (trains to predicting the most abundant structure) – You get 50% if you just predict everything to be a coil – Most methods get around 60% with this metric Correlation Coeficient • How correlated are the predictions for coils, helix and Beta-sheets to the real structures • This ignores what we really want to get to – If the real structure has 3 coils, do we predict 3 coils? • Segment overlap score (Sov) gives credit to how protein like the structure is, but it is correlated with Q3 Artificial Neural Network Predicts Structure at this point Danger • You may train the network on your training set, but it may not generalize to other data • Perhaps we should train several ANNs and then let them vote on the structure Profile network from HeiDelberg • family (alignment is used as input) instead of just the new sequence • On the first level, a window of length 13 around the residue is used • The window slides down the sequence, making a prediction for each residue • The input includes the frequency of amino acids occurring in each position in the multiple alignment (In the example, there are 5 sequences in the multiple alignment) • The second level takes these predictions from neural networks that are centered on neighboring proteins • The third level does a jury selection PHD Predicts 4 Predicts 5 Predicts 6 Fold recognition (structural profiles) • Attempts to find the best fit of a raw polypeptide sequence onto a library of known protein folds • A prediction of the secondary structure of the unknown is made and compared with the secondary structure of each member of the library of folds Threading • Takes the fold recognition process a step further: – Empirical-energy functions for residue pair interactions are used to mount the unknown onto the putative backbone in the best possible manner Fold recognition by threading Fold 1 Fold 2 Fold 3 Query sequence Compatibility scores Fold N CASP • http://www.predictioncenter.org/casp8/index. cgi SCOP • SCOP: Structural Classification of Proteins. • http://scop.mrc-lmb.cam.ac.uk/scop/ CATH • CATH: Protein Structure Classification • Class (C), Architecture (A), Topology (T) and Homologous superfamily (H)