Topic Modeling with MALLET

We'd like to test how Taylor Salo integrated MALLET into NeuroSynth, and whether that integration works in a docker container.

First, let's import some dependencies and text to work with.

For testing, we'll use an XML file separately downloaded from PubMed. In the spirit of NeuroSynth, we downloaded Tal Yarkoni's bibliography. Thanks, Tal!

In [1]:
from bs4 import BeautifulSoup
import pandas as pd

with open('../neurosynth/tests/data/yarkoni_pubmed.xml') as infile:
    xml_file =
soup = BeautifulSoup(xml_file, 'lxml')

    assert type(soup) == BeautifulSoup
except AssertionError:
    print('Check file type! Must be HTML or XML.')

titles = soup.find_all('articletitle')
abstracts = soup.find_all('abstract')

if len(titles) != len(abstracts):
    print('Warning: Some articles do not have abstracts on PubMed!')
    print('Only articles with complete data will be included.')
Warning: Some articles do not have abstracts on PubMed!
Only articles with complete data will be included.

Three articles do not have abstracts:

  1. Pain in the ACC?
  2. Introduction to the special issue on reliability and replication in cognitive and affective neuroscience research.
  3. Establishing homology between monkey and human brains.

Maybe because they're commentaries? We'll need to filter the results to only consider articles with abstracts. Then, import any matching articles into a pandas dataframe.

In [2]:
abstracts = []
pmids = []

articles = soup.find_all('pubmedarticle')
for a in articles:
    if a.find_all('abstract')!= []:
        # This is a little messy, but pulls out the
        # results in plain text without another loop.

df = pd.DataFrame({'pmid': pmids,
     'abstract': abstracts})

abstract pmid
0 Compassion is critical for societal wellbeing.... 27018610
1 Open access, open data, open source and other ... 27387362
2 The functional organization of human medial fr... 27307242
3 Social scientists often seek to demonstrate th... 27031707
4 Decades of animal and human neuroimaging resea... 26831091

We have a test dataset! Let's see how it plays with MALLET.

In [3]:
import os
import subprocess
import shutil
import sys
from neurosynth.analysis.reduce import topic_models

weights_df, keys_df = topic_models(df)
MALLET toolbox found!
Abstracts folder not found. Creating abstract files...
Generating topics...
topic_000 decision making neuroimaging gains demonstrate...
topic_001 social anxiety disorder generalized type givin...
topic_002 orthographic visual language widespread neighb...
topic_003 connectivity findings global found state lpfc ...
topic_004 data human provide coactivation brain map api ...