Mercurial > repos > recetox > matchms_formatter
view formatter.py @ 1:28a252f3b682 draft
planemo upload for repository https://github.com/RECETOX/galaxytools/tree/master/tools/matchms commit ca44513d50b29a4706e2a2db96c23ef6688b7c2d
author | recetox |
---|---|
date | Fri, 22 Jul 2022 16:49:20 +0000 |
parents | 60f34912b3de |
children | 574c6331e9db |
line wrap: on
line source
import click from pandas import DataFrame, read_csv def create_long_table(data: DataFrame, value_id: str) -> DataFrame: """Convert the table from compact into long format. See DataFrame.melt(...). Args: data (DataFrame): The data table to convert. value_id (str): The name to assign to the added column through conversion to long format. Returns: DataFrame: Table in long format. """ return data.transpose().melt(ignore_index=False, var_name='compound', value_name=value_id) def join_df(x: DataFrame, y: DataFrame, on=[], how="inner") -> DataFrame: """Shortcut functions to join to dataframes on columns and index Args: x (DataFrame): Table X y (DataFrame): Table Y on (list, optional): Columns on which to join. Defaults to []. how (str, optional): Join method, see DataFrame.join(...). Defaults to "inner". Returns: DataFrame: Joined dataframe. """ df_x = x.set_index([x.index] + on) df_y = y.set_index([y.index] + on) combined = df_x.join(df_y, how=how) return combined def get_top_k_matches(data: DataFrame, k: int) -> DataFrame: """Function to get top k matches from dataframe with scores. Args: data (DataFrame): A table with score column. k (int): Number of top scores to retrieve. Returns: DataFrame: Table containing only the top k best matches for each compound. """ return data.groupby(level=0, group_keys=False).apply(DataFrame.nlargest, n=k, columns=['score']) def filter_thresholds(data: DataFrame, t_score: float, t_matches: float) -> DataFrame: """Filter a dataframe with scores and matches to only contain values above specified thresholds. Args: data (DataFrame): Table to filter. t_score (float): Score threshold. t_matches (float): Matches threshold. Returns: DataFrame: Filtered dataframe. """ filtered = data[data['score'] > t_score] filtered = filtered[filtered['matches'] > t_matches] return filtered def load_data(scores_filename: str, matches_filename: str) -> DataFrame: """Load data from filenames and join on compound id. Args: scores_filename (str): Path to scores table. matches_filename (str): Path to matches table. Returns: DataFrame: Joined dataframe on compounds containing scores an matches in long format. """ matches = read_csv(matches_filename, sep=None, index_col=0) scores = read_csv(scores_filename, sep=None, index_col=0) scores_long = create_long_table(scores, 'score') matches_long = create_long_table(matches, 'matches') combined = join_df(matches_long, scores_long, on=['compound'], how='inner') return combined @click.group() @click.option('--sf', 'scores_filename', type=click.Path(exists=True), required=True) @click.option('--mf', 'matches_filename', type=click.Path(exists=True), required=True) @click.option('--o', 'output_filename', type=click.Path(writable=True), required=True) @click.pass_context def cli(ctx, scores_filename, matches_filename, output_filename): ctx.ensure_object(dict) ctx.obj['data'] = load_data(scores_filename, matches_filename) pass @cli.command() @click.option('--st', 'scores_threshold', type=float, required=True) @click.option('--mt', 'matches_threshold', type=float, required=True) @click.pass_context def get_thresholded_data(ctx, scores_threshold, matches_threshold): result = filter_thresholds(ctx.obj['data'], scores_threshold, matches_threshold) return result @cli.command() @click.option('--k', 'k', type=int, required=True) @click.pass_context def get_top_k_data(ctx, k): result = get_top_k_matches(ctx.obj['data'], k) return result @cli.resultcallback() def write_output(result: DataFrame, scores_filename, matches_filename, output_filename): input_file = read_csv(scores_filename, sep=None, iterator=True) sep = input_file._engine.data.dialect.delimiter result = result.reset_index().rename(columns={'level_0': 'query', 'compound': 'reference'}) result.to_csv(output_filename, sep=sep, index=False) if __name__ == '__main__': cli(obj={})