22 Statistics: compute Fisher exact

You can use pyranges to compute Fisher Exact quickly. To do so you need to have the package fisher installed.

import pyranges as pr
gr = pr.random(1000)
import numpy as np
n1 = np.random.randint(0, high=20, size=1000)
d2 = np.random.randint(0, high=20, size=1000)
n2 = np.random.randint(0, high=20, size=1000)
d1 = np.random.randint(0, high=20, size=1000)
fe = pr.stats.fisher_exact(n1, n2, d1, d2).head()

## /usr/share/miniconda/lib/python3.9/site-packages/pyranges-0.0.115-py3.9.egg/pyranges/statistics.py:163: RuntimeWarning: divide by zero encountered in true_divide
## /usr/share/miniconda/lib/python3.9/site-packages/pyranges-0.0.115-py3.9.egg/pyranges/statistics.py:163: RuntimeWarning: invalid value encountered in true_divide

print(fe)

##          OR         P     PLeft    PRight
## 0  0.972222  1.000000  0.620747  0.652569
## 1  0.333333  0.249881  0.145479  0.965552
## 2  1.909091  0.476752  0.891398  0.311049
## 3  0.333333  0.096384  0.051679  0.986081
## 4  0.836601  1.000000  0.500000  0.724388

You can get the false-discovery rate corrected fdr with pyranges.stats.fdr:

print(pr.stats.fdr(fe.P))

## 0    1.000000
## 1    0.624702
## 2    0.794587
## 3    0.481922
## 4    1.000000
## Name: P, dtype: float64