Refinement of domino method

[ChristopherMancuso]

This branch is for checking the domino implementation and improving it as needed. Some items that might be considered are

Questions

• Should any additional kwargs be added?
• Should parameter defaults be changed at all?
• In retain_relevant slices it finds them using connected components, might need to change to louvain for really dense networks at some point?
• In retain_relevant_slices, why did they not use number of pertubed nodes in cc instead of all pertubed nodes?
• In retain_relevant_slcies why does hypergeom use G_org for universe but only perturbed nodes found in G_modularity (this doesn’t account for perturbed nodes not in slices))?
• In prune_slice, since relevant_slices found with G_modularity, why look at them in the context of the full graph (G) here?
• In prune_slice and in run_pcst and in linear_threshold we can set the threshold and influence values of a node, do we need to do this ever?
• In prune_slice and in run_pcst, vertices_prizes seem to be mix of floats and booleans (i.e. Active value), is this OK?
• In general I don’t really get how in prune_slice the get_putatiave_modules is really working, should this be changed at all? In particular
  • Seem not great to just be looking at connected components
  • The if statement clause cur_modularity <= best_modularity + improvement_delta makes no sense to me and seems like if anything that should be a >= sign
• In the Louvain implementation of PyGenePlexus, is there a place we could do that in Domino?

Things I noticed in domino implementation

• Individual slices are never considered in domino, so the initial slices step just removes some network genes as slices are recombined right away
• Since linear_threshold only uses G_cc, then diffusion step only add scores for fast_pcst, but not nodes outside of G_cc. So does this mean all "propagated genes" come from finding relevant slices (and not diffusion) and then diffusion can just prune those.
• In prune_slice and in get_putative_modules, full_G and module_threshold are dropped from arguments. In original DOMINO there was code about finding sig_values for each sub-slice but those sig_values were never actually used
• In final_modules (and other places), modules that are below the min_clust_size parameters are included in the p-value adjustment calculations and they should probably be removed beforehand

[ChristopherMancuso]

The main functions to look at are

• geneplexus/cluster_input()
• _geneplexus/_generate_clusters()
• everything in _clustering/domino.py
• everything in _clustering/domino_utls.py

Some example code to run Geneplexus with clustering is

import geneplexus

### you will need to change the paths ###

# load in a gene set
# the code here loads the examples but it would be great if you tried some different ones
fp_pygp = "/Users/mancchri/Desktop/repos/PyGenePlexus/"
input_genes = geneplexus.util.read_gene_list(f"{fp_pygp}example/input_genes.txt")

# this is the path to the needed data. If you don't have it the downloader function should work (I can send code if needed)
# or you can get the data from here https://zenodo.org/records/14750555
fp_data = "/Users/mancchri/Desktop/CIDA_unsorted/Arjun/GenePlexusZoo_webserver/gpdata_prop/regular" # the good one right now

# set other variables
sp_trn = "Human" # needs to be a string of "Human", "Mouse", "Fly", "Zebrafish", "Worm", "Yeast"
gsc_trn = "Combined" # needs to be a STRING of either "GO", "Monarch", "Combined"
sp_res = ["Mouse", "Human", "Fly", "Yeast"] # can be a list or a single string (options same as sp_trn)
gsc_res = "Combined" # can be a list the length of the number or species or a string (if a string it will repeat the number of species for len(sp_res))
features = "SixSpeciesN2V"
net_type = "STRING" # needs to be a string, options "STRING", "BioGRID", "IMP"

# instansitate the class
gp = geneplexus.GenePlexus(file_loc = fp_data,
                           gsc_trn = gsc_trn,
                           gsc_res = gsc_res,
                           features = features,
                           net_type = net_type,
                           sp_trn = sp_trn,
                           sp_res = sp_res,
                           log_level="INFO")
# print(gp.sp_trn)
# print(gp.gsc_trn_original)
# print(gp.gsc_trn)
# print(gp.sp_res)
# print(gp.gsc_res_original)
# print(gp.gsc_res)
# print(gp.model_info["All-Genes"].results)
# print(list(gp.model_info))
# # run some functions
gp.load_genes(input_genes)
#### if you want to do louvain method uncomment below and comment out lines 49-55 ####
# lv_kwargs = {"louvain_max_size" : 70, "louvain_max_tries" : 3, "louvain_res" : 1, "louvain_seed" : 323}
# gp.cluster_input(
#     clust_method = "louvain",
#     clust_min_size = 5, # sets the minimum size of the clusters to keep. If you change also change gp.fit argument min_num_genes. This will be done automattically in the future
#     clust_weighted = True, # whether or not to use edge weights when making the clusters
#     clust_kwargs = lv_kwargs,
# )
dm_kwargs = {"domino_res" : 1, "domino_slice_thresh" : 0.3, "domino_n_steps" : 20, "domino_module_threshold" : 0.05, "domino_seed" : 123}
gp.cluster_input(
    clust_method = "domino",
    clust_min_size = 10,
    clust_weighted = True,
    clust_kwargs = dm_kwargs,
)
# print(gp.model_info["All-Genes"].model_genes)
# print(gp.model_info["Cluster-01"].model_genes)
# print(list(gp.model_info))
gp.fit(min_num_pos = 5)
# print(gp.model_info["All-Genes"].avgps)
# print(gp.model_info["Cluster-01"].avgps)
gp.predict()
# print(gp.model_info["All-Genes"].results["Fly-Combined"].df_probs)
# print(gp.model_info["Cluster-01"].results["Fly-Combined"].df_probs)
gp.make_sim_dfs()
# print(gp.model_info["All-Genes"].results["Fly-Combined"].df_sim)
# print(gp.model_info["Cluster-01"].results["Fly-Combined"].df_sim)
gp.make_small_edgelist()
# print(gp.model_info["All-Genes"].results["Fly-Combined"].df_edge_sym)
# print(gp.model_info["Cluster-01"].results["Fly-Combined"].df_edge_sym)
# gp.alter_validation_df()
# print(gp.df_convert_out_subset)
gp.save_class("/Users/mancchri/Desktop/save_test/mytest")