BPNet tf-Modisco report¶

url_dir = "http://mitra.stanford.edu/kundaje/avsec/chipnexus/oct-sox-nanog-klf/models/n_dil_layers=9/modisco/valid/new-hparams/plots/"
modisco_dir = "/users/avsec/workspace/basepair/data/processed/chipnexus/exp/models/oct-sox-nanog-klf/models/n_dil_layers=9/modisco/valid/new-hparams"

# Parameters
url_dir = "http://mitra.stanford.edu/kundaje/avsec/chipnexus/oct-sox-nanog-klf/models/n_dil_layers=9/modisco/clustered/weighted/2/plots/"
modisco_dir = "2"

from basepair.modisco import ModiscoResult
from basepair.config import get_data_dir
from basepair.utils import read_json
from basepair.plot.vdom import vdom_modisco
from kipoi.readers import HDF5Reader
import matplotlib.pyplot as plt
import numpy as np
import os
import pandas as pd
from plotnine import *

/users/avsec/bin/anaconda3/envs/chipnexus/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
  from ._conv import register_converters as _register_converters
Using TensorFlow backend.
/users/avsec/bin/anaconda3/envs/chipnexus/lib/python3.6/site-packages/statsmodels/compat/pandas.py:56: FutureWarning: The pandas.core.datetools module is deprecated and will be removed in a future version. Please use the pandas.tseries module instead.
  from pandas.core import datetools

mr = ModiscoResult(f"{modisco_dir}/modisco.h5")
mr.open()

# load the data
modisco_kwargs = read_json(os.path.join(modisco_dir, "kwargs.json"))
d = HDF5Reader(modisco_kwargs['imp_scores'])
d.open()

strand_dist_file = f"{modisco_dir}/strand_distances.h5"
if modisco_kwargs.get("ignore_strand_dist", False) and os.path.exists(strand_dist_file):
    included_samples = HDF5Reader.load(strand_dist_file)['included_samples']
else:
    included_samples = np.ones(d.f['inputs'].shape[:1], dtype=bool)
    
if modisco_kwargs.get("filter_npy", None) is not None:
    included_samples = np.load(modisco_kwargs['filter_npy']) * included_samples
    
    
id_hash = pd.DataFrame({"peak_id": d.f['/metadata/interval_from_task'][:][included_samples], 
                        "example_idx": np.arange(d.f['/metadata/interval_from_task'][included_samples].shape[0])})
tasks = list(d.f["targets"]["profile"].keys())

# get all seqlet instances
dfp = mr.seqlet_df_instances().rename(columns=dict(seqname="example_idx"))
dfp = pd.merge(dfp,id_hash, on="example_idx")

# row = example_idx
total_counts = pd.DataFrame({task: d.f[f"/targets/profile/{task}"][:][included_samples].sum(axis=-1).sum(axis=-1)
          for task in tasks
         })

len(mr.patterns())

79

# Number of metaclusters
len(mr.metaclusters())

37

Number of seqlets per pattern¶

mc_stat = mr.metacluster_stats()

ggplot(aes(x="pattern", y='n'), mc_stat) + geom_bar(stat='identity') + \
    facet_wrap("~metacluster", ncol=4, labeller='label_both') + \
    ylab("Number of seqlets") + theme_classic()

<ggplot: (-9223363296076078511)>

Zoom-into the 500 seqlet range¶

ggplot(aes(x="pattern", y='n'), mc_stat) + geom_bar(stat='identity') + \
    facet_wrap("~metacluster", ncol=4, labeller='label_both') + \
    ylab("Number of seqlets") + theme_classic() + coord_cartesian(ylim=[0, 500])

<ggplot: (8740781477026)>

Important tasks per metacluster¶

mcs_grouped = mc_stat.groupby("metacluster").n.agg(["count", "sum"]).reset_index()
fig, ax = plt.subplots(2, 1, sharex=False, figsize=(18,6), 
                       gridspec_kw={'height_ratios': [2,1]})
mcs_grouped.plot("metacluster", "count", 
                 label="# patterns per metacluster", style="o--", 
                 ax=ax[0], 
                 yticks=range(mcs_grouped['count'].max()+1),
                 xticks=range(38),
                 fontsize='large',
                 xlim=(-.5, len(mr.metaclusters()) - .5 ))
mcs_grouped.plot("metacluster", "sum", 
                 label="# seqlets per metacluster", 
                 style="o--", ax=ax[0], secondary_y=True)
ax[0].grid(linewidth=0.2)
mr.plot_metacluster_activity(ax[1], cbar=False)
ax[1].set_title("Importance score activity: Red = positive, Blue = negative");

vdom_modisco(mr, url_dir, total_counts, dfp, is_open=True, trim_frac=0.08, letter_width=0.15, height=0.5)

print("Metaclusters heatmap")
import seaborn as sns
activity_patterns = np.array(mr.f.f['metaclustering_results']['attribute_vectors'])[
                    np.array(
        [x[0] for x in sorted(
                enumerate(mr.f.f['metaclustering_results']['metacluster_indices']),
               key=lambda x: x[1])])]
sns.heatmap(activity_patterns, center=0);

Metaclusters heatmap