Link to results

Results

import os
import sys
sys.path.append(os.path.abspath("/users/amtseng/tfmodisco/src/"))
from util import figure_to_vdom_image, create_motif_similarity_matrix, aggregate_motifs, aggregate_motifs_from_inds, purine_rich_motif
import plot.viz_sequence as viz_sequence
import numpy as np
import h5py
import matplotlib.pyplot as plt
import vdom.helpers as vdomh
from IPython.display import display

Define constants and paths

# Define parameters/fetch arguments
in_motif_file = os.environ["TFM_IN_MOTIF_FILE"]
out_motif_file = os.environ["TFM_OUT_MOTIF_FILE"]

print("Input motif file: %s" % in_motif_file)
print("Output motif file: %s" % out_motif_file)

Input motif file: /users/amtseng/tfmodisco/results/motifs/tfmodisco//E2F6_tfmodisco_motifs.h5
Output motif file: /users/amtseng/tfmodisco/results/motifs/tfmodisco//E2F6_tfmodisco_cpmerged_motifs.h5

min_match_sim = 0.9

os.makedirs(os.path.dirname(out_motif_file), exist_ok=True)

Helper functions

def import_motifs_from_hdf5_group(h5_group):
    """
    Imports a set of motifs from an open HDF5 group.
    The HDF5 group must be structured as follows:
        count:
            0_0:
                cwm_trimmed
                ...
        profile:
            0_0:
                cwm_trimmed
                ...
    Returns a dictionary of the trimmed CWMs matching the
    structure of the group.
    Motifs are flipped to the purine-rich version.
    """
    motifs = {}
    for key in ("count", "profile"):
        motifs[key] = {}
        for motif_key in h5_group[key]:
            cwm = h5_group[key][motif_key]["cwm_trimmed"][:]
            motifs[key][motif_key] = purine_rich_motif(cwm)
        
    return motifs

def write_merged_motifs_to_hdf5_group(in_h5_group, out_h5_group, matched_keys):
    """
    Writes a set of motifs from an open HDF5 group.
    The input HDF5 group must be structured as follows:
        count:
            0_0:
                cwm_trimmed
                ...
        profile:
            0_0:
                cwm_trimmed
                ...
    The output HDF5 group will be stuctured as follows:
        C0_0:P0_1:
            cwm_trimmed
            ...
    `matched_keys` is a list of pairs of keys, denoting count
    and profile keys that are matched together, respectively.
    Motifs are flipped to the purine-rich version.
    """
    count_keys = list(in_h5_group["count"].keys())
    profile_keys = list(in_h5_group["profile"].keys())

    matched_count, matched_profile = (list(zip(*matched_keys))) if matched_keys else ([], [])
    
    unmatched_count = [key for key in count_keys if key not in matched_count]
    unmatched_profile = [key for key in profile_keys if key not in matched_profile]
    
    for count_key, profile_key in matched_keys:
        group = out_h5_group.create_group("C%s:P%s" % (count_key, profile_key))
        # First aggregate the CWMs to get the indices
        for length in ("full", "trimmed"):
            _, (const_inds, agg_inds) = aggregate_motifs([
                purine_rich_motif(in_h5_group["count"][count_key]["cwm_%s" % length][:]),
                purine_rich_motif(in_h5_group["profile"][profile_key]["cwm_%s" % length][:])
            ], return_inds=True)
            for motif_prefix in ["pfm", "cwm", "hcwm"]:
                motif_type = "%s_%s" % (motif_prefix, length)
                agg = aggregate_motifs_from_inds([
                    purine_rich_motif(in_h5_group["count"][count_key][motif_type][:]),
                    purine_rich_motif(in_h5_group["profile"][profile_key][motif_type][:])
                ], const_inds, agg_inds)
                group.create_dataset(motif_type, data=agg, compression="gzip")
    
    for count_key in unmatched_count:
        group = out_h5_group.create_group("C%s" % count_key)
        for motif_type in ["pfm_full", "cwm_full", "hcwm_full", "pfm_trimmed", "cwm_trimmed", "hcwm_trimmed"]:
            motif = purine_rich_motif(in_h5_group["count"][count_key][motif_type][:])
            group.create_dataset(motif_type, data=motif, compression="gzip")
    for profile_key in unmatched_profile:
        group = out_h5_group.create_group("P%s" % profile_key)
        for motif_type in ["pfm_full", "cwm_full", "hcwm_full", "pfm_trimmed", "cwm_trimmed", "hcwm_trimmed"]:
            motif = purine_rich_motif(in_h5_group["profile"][profile_key][motif_type][:])
            group.create_dataset(motif_type, data=motif, compression="gzip")

def compute_motif_pairs(motifs_a, motifs_b):
    """
    For a list of motifs `motifs_a` and another list of motifs `motifs_b`,
    computes the similarity between all pairs and pairs up the closest
    motifs between A and B.
    This is done greedily, where the two motifs with the biggest similarity
    in the matrix are paired up, and then removed from the matrix.
    Returns a list of triplets (A_i, B_i, s_i), where A_i and B_i are the
    indices of motifs in `motifs_a` and `motifs_b` that have been matched
    up, and `s_i` is the similarity between them. The triplets are ordered
    by decreasing `s_i`. Note that if the lengths of the lists are not the
    same, then some motifs will not be matched up.
    """
    num_a, num_b = len(motifs_a), len(motifs_b)
    
    # Compute the similarity matrix
    sim_matrix = create_motif_similarity_matrix(motifs_a, motifs_2=motifs_b, show_progress=False)
    
    a_inds, b_inds = np.arange(num_a), np.arange(num_b)
    matches = []
    for _ in range(min(num_a, num_b)):
        a_i, b_i = np.unravel_index(np.argmax(sim_matrix), sim_matrix.shape)
        sim = sim_matrix[a_i, b_i]
        
        matches.append((a_inds[a_i], b_inds[b_i], sim))
        
        a_inds = np.delete(a_inds, a_i)
        b_inds = np.delete(b_inds, b_i)
        sim_matrix = np.delete(np.delete(sim_matrix, a_i, axis=0), b_i, axis=1)
        
    return matches

def plot_matches(count_keys, count_motifs, profile_keys, profile_motifs, matches):
    """
    Show matched and leftover motifs.
    """
    display(vdomh.h4("Matched motifs"))
    colgroup = vdomh.colgroup(
        vdomh.col(style={"width": "5"}),
        vdomh.col(style={"width": "5"}),
        vdomh.col(style={"width": "45%"}),
        vdomh.col(style={"width": "45%"}),
    )
    header = vdomh.thead(
        vdomh.tr(
            vdomh.th("Merged ID", style={"text-align": "center"}),
            vdomh.th("Similarity", style={"text-align": "center"}),
            vdomh.th("Aggregate motif", style={"text-align": "center"}),
            vdomh.th("Constituent motifs", style={"text-align": "center"})
        )
    )

    rows = []
    for count_i, profile_i, sim in matches:
        constituents = [count_motifs[count_i], profile_motifs[profile_i]]
        consensus = aggregate_motifs([count_motifs[count_i], profile_motifs[profile_i]])

        merged_id = "C%s:P%s" % (count_keys[count_i], profile_keys[profile_i])

        agg_fig = viz_sequence.plot_weights(consensus, figsize=(20, 4), return_fig=True)
        agg_fig.tight_layout()
        const_figs = []
        for motif in constituents:
            fig = viz_sequence.plot_weights(motif, figsize=(20, 4), return_fig=True)
            fig.tight_layout()
            const_figs.append(figure_to_vdom_image(fig))

        rows.append(vdomh.tr(
            vdomh.td(merged_id),
            vdomh.td("%.4f" % sim),
            vdomh.td(figure_to_vdom_image(agg_fig)),
            vdomh.td(*const_figs)
        ))
    display(vdomh.table(colgroup, header, vdomh.tbody(*rows)))
    plt.close("all")

    display(vdomh.h4("Unmatched motifs"))
    colgroup = vdomh.colgroup(
        vdomh.col(style={"width": "5%"}),
        vdomh.col(style={"width": "45%"}),
        vdomh.col(style={"width": "5%"}),
        vdomh.col(style={"width": "45%"}),
    )
    header = vdomh.thead(
        vdomh.tr(
            vdomh.th("Count ID", style={"text-align": "center"}),
            vdomh.th("Count motif", style={"text-align": "center"}),
            vdomh.th("Profile ID", style={"text-align": "center"}),
            vdomh.th("Profile motif", style={"text-align": "center"})
        )
    )

    matched_count = [trip[0] for trip in matches]
    unmatched_count = [i for i in range(len(count_motifs)) if i not in matched_count]
    matched_profile = [trip[1] for trip in matches]
    unmatched_profile = [i for i in range(len(profile_motifs)) if i not in matched_profile]
    rows = []
    for i in range(max(len(unmatched_count), len(unmatched_profile))):
        row = []
        if i < len(unmatched_count):
            fig = viz_sequence.plot_weights(
                count_motifs[unmatched_count[i]], figsize=(20, 4), return_fig=True
            )
            fig.tight_layout()
            row.extend([
                vdomh.td("C%s" % count_keys[unmatched_count[i]]),
                vdomh.td(figure_to_vdom_image(fig))
            ])
        else:
            row.extend([vdomh.td(), vdomh.td()])
        if i < len(unmatched_profile):
            fig = viz_sequence.plot_weights(
                profile_motifs[unmatched_profile[i]], figsize=(20, 4), return_fig=True
            )
            fig.tight_layout()
            row.extend([
                vdomh.td("P%s" % profile_keys[unmatched_profile[i]]),
                vdomh.td(figure_to_vdom_image(fig))
            ])
        else:
            row.extend([vdomh.td(), vdomh.td()])
        rows.append(vdomh.tr(*row))
    display(vdomh.table(colgroup, header, vdomh.tbody(*rows)))
    plt.close("all")

motif_key_sorter = lambda k: (int(k.split("_")[0]), int(k.split("_")[1]))

Consolidate count/profile head motifs for all conditions

def consolidate(in_h5_group, out_h5_group):
    """
    Performs consolidation, reading motifs from the in group and
    writing to the out group.
    """
    # Import CWMs
    motifs = import_motifs_from_hdf5_group(in_h5_group)
    count_keys = sorted(motifs["count"].keys(), key=motif_key_sorter)
    profile_keys = sorted(motifs["profile"].keys(), key=motif_key_sorter)
    count_motifs = [motifs["count"][k] for k in count_keys]
    profile_motifs = [motifs["profile"][k] for k in profile_keys]
    matches = compute_motif_pairs(
        count_motifs, profile_motifs
    )

    # Plot distribution of match similarities
    fig, ax = plt.subplots(figsize=(10, 4))
    ax.hist([trip[2] for trip in matches], bins=10)
    ax.set_title("Histogram of match distances")
    ax.set_xlabel("Similarity")
    plt.show()

    # Restrict to matches of sufficient similarity
    matches = [match for match in matches if match[2] >= min_match_sim]

    # Plot matches themselves
    plot_matches(count_keys, count_motifs, profile_keys, profile_motifs, matches)

    # Write the consolidated motifs and unconsolidated motifs
    matched_keys = [(count_keys[trip[0]], profile_keys[trip[1]]) for trip in matches]
    write_merged_motifs_to_hdf5_group(in_h5_group, out_h5_group, matched_keys)

with h5py.File(in_motif_file, "r") as f, h5py.File(out_motif_file, "w") as g:
    # Multi-task, all 10 folds
    f_mt = f["multitask"]
    g_mt = g.create_group("multitask")
    for fold in f_mt.keys():
        display(vdomh.h3("Multi-task %s" % fold))
        
        f_mt_fold = f_mt[fold]
        g_mt_fold = g_mt.create_group(fold)
        
        consolidate(f_mt_fold, g_mt_fold)
        
    f_st = f["singletask"]
    g_st = g.create_group("singletask")
    
    # Single-task, all 10 folds for all tasks
    for task in f_st.keys():
        f_st_task = f_st[task]
        g_st_task = g_st.create_group(task)
        
        for fold in f_st_task.keys():
            display(vdomh.h3("Single-task %s %s" % (task, fold)))

            f_st_task_fold = f_st_task[fold]
            g_st_task_fold = g_st_task.create_group(fold)

            consolidate(f_st_task_fold, g_st_task_fold)
            
    # Multi-task fine-tuned, all tasks
    f_mtft = f["multitask_finetune"]
    g_mtft = g.create_group("multitask_finetune")
    
    for task in f_mtft.keys():
        display(vdomh.h3("Multi-task fine-tune %s" % task))
        f_mtft_task = f_mtft[task]
        g_mtft_task = g_mtft.create_group(task)

        consolidate(f_mtft_task, g_mtft_task)
        
    # Single-task fine-tuned, all tasks
    f_stft = f["singletask_finetune"]
    g_stft = g.create_group("singletask_finetune")
    
    for task in f_stft.keys():
        display(vdomh.h3("Single-task fine-tune %s" % task))
        f_stft_task = f_stft[task]
        g_stft_task = g_stft.create_group(task)

        consolidate(f_stft_task, g_stft_task)

Multi-task fold_1

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9913
C0_1:P0_1	0.9907
C0_3:P0_4	0.9677
C0_2:P0_5	0.9090

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_4		P0_2
C0_5		P0_3
		P0_6
		P0_7
		P0_8
		P0_9
		P0_10

Multi-task fold_10

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9973
C0_1:P0_1	0.9934
C0_3:P0_3	0.9662
C0_2:P0_6	0.9434

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_4		P0_2
C0_5		P0_4
C0_6		P0_5
C0_7		P0_7
		P0_8
		P0_9
		P0_10
		P0_11

Multi-task fold_2

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9908
C0_5:P0_6	0.9856
C0_1:P0_1	0.9684

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_2		P0_2
C0_3		P0_3
C0_4		P0_4
		P0_5
		P0_7
		P0_8
		P0_9
		P1_0

Multi-task fold_3

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9938
C0_2:P0_2	0.9917
C0_3:P0_4	0.9850
C0_4:P0_3	0.9520
C0_1:P0_1	0.9309
C0_5:P0_8	0.9250

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_6		P0_5
C0_7		P0_6
C0_8		P0_7
		P0_9
		P0_10

Multi-task fold_4

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9972
C0_2:P0_3	0.9295
C0_1:P0_1	0.9210

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_3		P0_2
C0_4		P0_4
C0_5		P0_5
		P1_0
		P1_1

Multi-task fold_5

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9973
C0_1:P0_1	0.9835
C0_3:P0_3	0.9398

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_2		P0_2
C0_4		P0_4
C0_5		P0_5
C0_6		P0_6
		P0_7

Multi-task fold_6

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_5:P0_4	0.9898
C0_4:P0_5	0.9800
C0_1:P0_1	0.9790
C0_3:P0_11	0.9534
C0_0:P0_0	0.9488

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_2		P0_2
		P0_3
		P0_6
		P0_7
		P0_8
		P0_9
		P0_10
		P0_12
		P0_13

Multi-task fold_7

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9728
C0_2:P0_5	0.9630
C0_3:P0_4	0.9516
C0_1:P0_1	0.9029

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
		P0_2
		P0_3
		P0_6
		P0_7
		P0_8
		P0_9
		P0_10
		P1_0

Multi-task fold_8

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9912
C0_3:P0_3	0.9832
C0_4:P0_4	0.9739
C0_6:P0_10	0.9698
C0_1:P0_1	0.9564

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_2		P0_2
C0_5		P0_5
		P0_6
		P0_7
		P0_8
		P0_9
		P0_11
		P0_12

Multi-task fold_9

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9971
C0_2:P0_2	0.9872
C0_3:P0_3	0.9558
C0_1:P0_1	0.9306

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_4		P0_4
		P0_5
		P1_0

Single-task task_0 fold_1

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_1:P0_1	0.9973
C0_2:P0_2	0.9851
C0_0:P0_0	0.9843

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_3		P0_3
		P0_4

Single-task task_0 fold_10

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9883
C0_5:P0_9	0.9809
C0_3:P0_2	0.9696
C0_1:P0_1	0.9421

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_2		P0_3
C0_4		P0_4
C0_6		P0_5
		P0_6
		P0_7
		P0_8
		P0_10
		P0_11

Single-task task_0 fold_2

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_3:P0_3	0.9928
C0_2:P0_2	0.9794
C0_0:P0_0	0.9757
C0_4:P0_4	0.9740
C0_1:P0_1	0.9702
C0_7:P0_6	0.9191

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_5		P0_5
C0_6		P0_7
		P0_8

Single-task task_0 fold_3

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_1:P0_1	0.9913
C0_0:P0_0	0.9885
C0_2:P0_2	0.9635
C0_3:P0_3	0.9631

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
		P0_4
		P0_5
		P1_0
		P1_1

Single-task task_0 fold_4

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9947
C0_1:P0_1	0.9942
C0_2:P0_2	0.9604

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_3		P0_3
C0_4		P0_4
		P0_5
		P0_6
		P0_7
		P0_8
		P0_9
		P0_10
		P0_11
		P0_12
		P0_13
		P0_14
		P0_15

Single-task task_0 fold_5

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_1:P0_1	0.9943
C0_2:P0_2	0.9645
C0_0:P0_0	0.9270

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
		P0_3
		P0_4
		P0_5
		P0_6

Single-task task_0 fold_6

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_1:P0_1	0.9980
C0_0:P0_0	0.9968
C0_2:P0_6	0.9400

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_3		P0_2
C0_4		P0_3
C0_5		P0_4
		P0_5
		P0_7
		P0_8

Single-task task_0 fold_7

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9962
C0_3:P0_4	0.9923
C0_1:P0_2	0.9914
C0_4:P0_3	0.9888
C0_2:P0_1	0.9108

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_5		P0_5
C0_6

Single-task task_0 fold_8

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9969
C0_1:P0_1	0.9947
C0_4:P0_4	0.9800
C0_3:P0_3	0.9691
C0_2:P0_5	0.9347

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_5		P0_2
		P0_6
		P0_7
		P0_8

Single-task task_0 fold_9

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9955
C0_1:P0_2	0.9841
C0_2:P0_1	0.9772

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
		P0_3
		P1_0

Single-task task_1 fold_1

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9964
C0_1:P0_2	0.9947

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_2		P0_1
C0_3		P0_3
		P0_4
		P0_5
		P0_6

Single-task task_1 fold_10

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9951
C0_1:P0_1	0.9922
C0_2:P0_2	0.9377

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_3		P0_3
		P0_4
		P0_5
		P0_6

Single-task task_1 fold_2

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9903
C0_3:P0_1	0.9781

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_1		P0_2
C0_2		P0_3
		P0_4
		P0_5
		P0_6

Single-task task_1 fold_3

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9927
C0_2:P0_1	0.9796
C0_1:P0_5	0.9714
C0_5:P0_3	0.9423

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_3		P0_2
C0_4		P0_4
		P0_6
		P0_7
		P0_8
		P0_9

Single-task task_1 fold_4

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9918
C0_1:P0_2	0.9831
C0_3:P0_3	0.9708
C0_2:P0_1	0.9044

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
		P0_4
		P0_5

Single-task task_1 fold_5

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_2:P0_1	0.9907
C0_3:P0_2	0.9904
C0_1:P0_3	0.9792
C0_0:P0_0	0.9766

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
		P0_4
		P0_5
		P0_6
		P0_7
		P0_8
		P0_9
		P0_10

Single-task task_1 fold_6

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_1	0.9955
C0_2:P0_2	0.9942
C0_1:P0_4	0.9403
C0_4:P0_0	0.9243

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_3		P0_3
C0_5		P0_5
		P0_6
		P0_7
		P0_8
		P0_9
		P0_10
		P1_0

Single-task task_1 fold_7

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9941
C0_3:P0_3	0.9895
C0_1:P0_1	0.9879
C0_2:P0_5	0.9443

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_4		P0_2
		P0_4
		P0_6
		P0_7

Single-task task_1 fold_8

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9838
C0_4:P0_2	0.9740
C0_3:P0_1	0.9551

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_1		P0_3
C0_2		P0_4
C0_5		P0_5
C0_6		P0_6
		P0_7
		P0_8
		P0_9

Single-task task_1 fold_9

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9992
C0_2:P0_1	0.9899

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_1		P0_2
C0_3		P0_3
C0_4		P0_4
		P0_5

Multi-task fine-tune task_0

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9977
C0_2:P0_2	0.9945
C0_1:P0_1	0.9825
C0_3:P0_3	0.9671

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_4		P0_4
C0_5		P0_5
C0_6		P0_6
		P0_7
		P0_8
		P0_9

Multi-task fine-tune task_1

Matched motifs

/users/amtseng/tfmodisco/src/plot/viz_sequence.py:152: RuntimeWarning: More than 20 figures have been opened. Figures created through the pyplot interface (`matplotlib.pyplot.figure`) are retained until explicitly closed and may consume too much memory. (To control this warning, see the rcParam `figure.max_open_warning`).
  fig = plt.figure(figsize=figsize)

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9866
C0_3:P0_3	0.9764
C0_1:P0_1	0.9729
C0_2:P0_7	0.9437
C0_6:P0_2	0.9433
C0_4:P0_5	0.9096
C0_7:P0_11	0.9027

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_5		P0_4
		P0_6
		P0_8
		P0_9
		P0_10
		P0_12

Multi-task fine-tune task_agg

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9928
C0_1:P0_1	0.9863
C0_3:P0_5	0.9570
C0_2:P0_2	0.9499
C0_5:P0_8	0.9076
C0_4:P0_3	0.9011

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
		P0_4
		P0_6
		P0_7
		P0_9
		P0_10
		P0_11

Single-task fine-tune task_0

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9872
C0_4:P0_9	0.9676
C0_3:P0_2	0.9515
C0_1:P0_1	0.9165

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_2		P0_3
		P0_4
		P0_5
		P0_6
		P0_7
		P0_8
		P0_10
		P0_11

Single-task fine-tune task_1

Matched motifs

Merged ID	Similarity	Aggregate motif	Constituent motifs
C0_0:P0_0	0.9981
C0_2:P0_2	0.9943

Unmatched motifs

Count ID	Count motif	Profile ID	Profile motif
C0_1		P0_1
		P0_3
		P0_4
		P0_5