Update documentation

[liamgd]

The documentation should be updated as well as the function docstrings (particularly for regression full). Unnecessary and hard to maintain portions of the README should be removed as changes to the code in the future are not updated in the documentation without manually editing. The command line interface does not need extensive documentation, as the --help argument provides most of this information. Instead, a concise overview of commands would be more useful.

[liamgd]

Updated in 2143306.

[liamgd]

Added regression_full MLR docstring in 2a0d3f5.

    Calculates the multiple linear regression of the input dataframes M,
    G, and C, being methylation beta values, gene expression values, and
    covariates using torch. This is done for every pair of methylation
    id and gene id. The regression formula is G ~ M + C1 + C2 + ...

    The p-values are the Student's T CDF function evaluated on the t
    statistic for each regression. Torch does not currently support the
    Student's T CDF function or any function that would help to
    implement it in python. Instead, the normal distribution CDF is used
    as an approximation of the Student's T CDF. For more degrees of
    freedom (ie. more samples, fewer covariates), this approximation
    is more accurate.

    M_annot and G_annot are annotation files that provide the positions
    of each methylation and gene expression id. They are optional and
    only required for region filtration.

    Region filtration filters the input by the distance between the
    methylation id and the gene expression id for each regression. Cis
    filtration only allows regressions where the ids are within a
    specified distance, the window, on the same chromosome. Distal
    analyses only allow regressions with ids with a distance greater
    than the window on the same chromosome. Trans analyses only allow
    regressions with methylation and gene expression ids on the same
    chromosome. The last region filtration mode, all, does not filter
    by region.

    P-value filtration filters the output of the regression by only
    including regressions with a p-value below p_thresh.

    For larger inputs, one may encounter memory limits. If this is the
    case, there are two ways of chunking the input data to avoid these
    limits: methylation chunking and gene expression chunking. Specify
    the number of meth_loci_per_chunk and gene_loci_per_chunk. Gene
    expression chunking is less detrimental to performance than
    methylation chunking, but both shuold be avoided as they sacrifice
    parallelization and speed. Both chunking methods are optional, and
    they can be combined together as well. Use the chunks command to
    estimate how many gene_loci_per_chunk to use for given settings. If
    no chunking is used, the output dataframe is returned. If chunking
    is used, chunks are saved to output files in output_dir.

    The methylation_only boolean option which defaults to true
    determines whether only methylation results should be saved in the
    output. If false, the intercept, methylation, and covariate results
    are saved. Regardless of this value, the intercept and covariates
    are used in the regression calculation. This parameter only affects
    the inclusion of output data.

    The p_only boolean option which defaults to false determines whether
    to include the estimate, standard error, Student's T statistic, and
    p-value (if false) or just the p-value (if true) for a faster saving
    time and lower output size.

    The file_format parameter is a string that determines the file name
    of each chunk saved in output_dir based on a formatting string with
    the parameters meth_chunk for the methylation chunk number and
    gene_chunk for the gene expression chunk number. The string is
    formatted using:
        file_format.format(
            meth_chunk=meth_index_str,
            gene_chunk=gene_index_str,
        )