FSWeight and Other Predictions Version 2.2 ============================== USAGE: predict.pl -i interactions \ [-s annotscheme] [-a annotations] [-o outputfile] \ [-p action] [-f folds] [-x informative] [-k] \ [-m prediction method] [-w weighting scheme] \ [-t protein list] options: General Parameters: -o Output File (Default = output.txt) This specifies an output file for which results are printed. -p Action (Default = p) p = prediction of novel annotations This will print predictions for all unannotated proteins to the specified output file. The format of the output is as follows: Protein Name\tGO Term\tScore E.g.: SGD_S000001924 0016481 0.121650140097154 c = cross validation This will perform a N-Fold cross validation for predictions using Informative GO Terms. N is specified using the -f specifier. When N is set to 1 (Default), LOOCV is performed. Informative threshold is set using the -x specifier. (Default = 30) Predictions for each fold will be printed to the specified output file. Cross Validation results (Precision-Recall analysis and ROC analysis) are printed to the standard output. The format of the output is as follows: a) Precision-Recall Analysis: 1) First Line: No. of Annotations\tPredicted Annotations\tNo. of proteins predicted E.g.: 5299 60079 4889 2) Subsequent Lines: Score Threshold\tPredictions\tRecall\tPrecision E.g.: 0.743805825614686 112 0.0200037742970372 0.946428571428571 b) ROC (Receiver Operating Characteristics) Analysis GO Namespace|GO Level|Description|Parent GO Terms\tGO Term\tROC E.g.: biological_process|6|cell cycle checkpoint|0000074 0000075 0.956277508936605 For more information on Cross-Validation, GO Level, Informative GO Terms and validation measures, please refer to [1] & [2] w = weight interactions This will print the topological weight of network edges to the specified output file. Weighting Scheme is specified using the -w specifier. (Default = FSWEIGHT) The format of the output is as follows: Protein A\tProtein B\tType\tWeight E.g.: SGD_S000005552 SGD_S000005939 L1 0.107143836623493 Functional Annotations: -s Annotation Scheme This is a flat text file describing GO Terms. The format of the file is as follows: GO ID\tGO Namespace|GO Level|Description|Parent GO Terms E.g.: 0051177 biological_process|8|meiotic sister chromatid cohesion|0007062,0007127,0045132 Processed GO Scheme files can be found at http://srs2.bic.nus.edu.sg/~kenny/integration/ This input is necessary when p or c are selected for the -p specifier. Non-GO annotation schemes are also supported. See TIPS below. -a Annotations This is a flat file describing GO Annotations. The format of the file is as follows: Protein Name|GO Term| E.g.: UniProt_O77783|0050508| A GO Annotation file can be found at http://srs2.bic.nus.edu.sg/~kenny/integration/ This input is necessary when p or c are selected for the -p specifier. Non-GO annotation schemes are also supported. See TIPS below. -v Vague List This is a comma-delimited list of vague annotation IDs which will be ignored during prediction or cross validation. Default list is 0000004,0005554,0008372,0003674,0005575,0008150 for GO Annotations If no vague terms are defined, set -v 0. -r Minimum Meaningful Annotation Level This is the minimum annotation level (for a hierarchical annotation scheme) in which the sharing of function is considered meaningful. The default level is 4 for GO annotations. For non-GO annotation scheme, -r 1 usually suffices. TIP: If an annotation scheme other than Gene ontology is used: a) Non-Hierarchical annotation scheme i) Supply annotation scheme file in the following format: ID\tName ii) Set options -v 0 and -r 1 b) Hierarchical annotation scheme i) Supply annotation scheme file in the following format: GO ID\t|Level|Description|Parent Terms ii) Propagate all annotations so that true path annotation is ensured. iii) Set options -v based on vague terms in the scheme e.g. 799, 96, 98, 99 for MPS FunCat 2.0 iv) Set options -r 1 Protein Binary Pairs: -i Interactions This is a flat file describing binary protein relationship (e.g. interactions) The format of the file is as follows: Protein A\tProtein B\tDatasource Name\tWeight E.g.: SGD_S000001474 SGD_S000005922 L1 0.0520662450872069 Weight is optional and is only used for the PROB_COMBINE prediction method. Cross Validation: -f Folds (Default = 1) This specifies the number of folds for cross validation. (Only applies when -p c is set) Defaults to 1, which will perform a Leave-One-Out cross validation (LOOCV). -x Threshold for Informative GO Terms (Default = 30) The minimum number of proteins annotated to a GO Term to define Informative GO Terms. See [1] & [2] for more details. -k Ignore isolated proteins (no neighbours) (Default = Off) If set, isolated proteins (i.e. not involved in any interactions or relationship specified by -i) will be ignored when computing informative GO Terms and during cross validation. Function Prediction: -m Prediction Method (MAJORITY_VOTE, CHI_SQUARE, WEIGHTED_AVG, PROB_COMBINE) (Default = PROB_COMBINE) Specify prediction method to use. MAJORITY_VOTE - Neighbour Counting Method [3] CHI_SQUARE - Chi-Square Method [4] WEIGHTED_AVG - Weighted Averaging [1] & [2] PROB_COMBINE - Data Fusion Methods [5] -w Weighting Scheme (CD_DIST, GEOMETRIC, FSWEIGHT) (Default = FSWEIGHT) Specify topological weight for weighting edges. CD_DIST - 1 - Czekanowski-Dice distance [6] GEOMETRIC - Geometric Distance [7] FSWEIGHT - FS-Weight [1] & [2] -t Protein List File (Perform prediction/validation only on selected proteins) List of proteins to predict or validate. If this is not specified, all proteins will be used. Credits: This program is implemented by CHUA Hon Nian, and is supported in part by an NGS scholarship, and URC grant "R-252-000-274-112: Graph-Based Protein Function Prediction". If you use this program, please cite Refs 1, 2, 5 below. References: [1] H. N. Chua, W.K. Sung, L. Wong. (2006) Exploiting indirect neighbours and topological weight to predict protein function from protein-protein interactions. Bioinformatics, 22:1623-1630. [2] H. N. Chua, W.K. Sung, L. Wong. (2007) Using Indirect Protein Interactions for the Prediction of Gene Ontology Functions. BMC Bioinformatics, 8(Suppl 4):S8. [3] B. Schwikowski, P. Uetz, S. Fields. (2000) A network of interacting proteins in yeast. Nat Biotechnol, 18:1257-1261. [4] H. Hishigaki, K. Nakai, T. Ono, A. Tanigami, T. Takagi. (2001) Assessment of prediction accuracy of protein function from protein-protein interaction data. Yeast, 18:525-531. [5] H. N. Chua, W.K. Sung, L. Wong. (2007) An efficient strategy for extensive integration of diverse biological data for protein function prediction. Bioinformatics, 23(24):3364-3373. [6] C. Brun, F. Chevenet, D. Martin, J. Wojcik, A. Guénoche, B. Jacq. (2003) Functional classification of proteins for the prediction of cellular function from a protein-protein interaction network. Genome Biol, 5:R6. [7] D.S. Goldberg and F.P. Roth. (2002) Assessing experimentally derived interactions in a small world. Proc Natl Acad Sci U S A 2003 Apr 15 100(8):4372-6.