nisthal dataset

# Standard imports
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

# Special imports
import mavenn

Summary

The DMS dataset from Nisthal et al. (2019). The authors used a high-throughput protein stability assay to measure folding energies for single-mutant variants of GB1. Column 'x' list variant GB1 sequences (positions 2-56). Column 'y' lists the Gibbs free energy of folding (i.e., \(\Delta G_F\)) in units of kcal/mol; lower energy values correspond to increased protein stability. Sequences are not divided into training, validation, and test sets because this dataset is only used for validation in Tareen et al. (2021).

Name: 'nisthal'

Reference: Nisthal A, Wang CY, Ary ML, Mayo SL. Protein stability engineering insights revealed by domain-wide comprehensive mutagenesis. Proc Natl Acad Sci USA 116:16367–16377 (2019).

mavenn.load_example_dataset('nisthal')

	x	name	y
0	AYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	T02A	0.4704
1	DYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	T02D	0.5538
2	EYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	T02E	-0.1299
3	FYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	T02F	-0.3008
4	GYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	T02G	0.6680
...	...	...	...
913	TYTLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	K04T	-0.4815
914	TYVLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	K04V	0.2696
915	TYYLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	K04Y	-0.8246
916	VYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	T02V	-1.3090
917	YYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	T02Y	-0.1476

918 rows × 3 columns

Preprocessing

First we load and preview the raw dataset published by Nisthal et al. (2019)

raw_data_file = '../../mavenn/examples/datasets/raw/nisthal_raw.csv'
raw_df = pd.read_csv(raw_data_file)
raw_df

	Sequence	Description	Ligand	Data	Units	Assay/Protocol
0	ATYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYD...	M01A	NaN	NaN	kcal/mol	ddG(deepseq)_Olson
1	ATYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYD...	M01A	NaN	NaN	kcal/mol	ddG_lit_fromOlson
2	ATYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYD...	M01A	NaN	-1.777	kcal/mol·M	m-value
3	ATYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYD...	M01A	NaN	-0.635	kcal/mol	FullMin
4	ATYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYD...	M01A	NaN	-0.510	kcal/mol	Rosetta SomeMin_ddG
...	...	...	...	...	...	...
18856	YTYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYD...	M01Y	NaN	0.512	kcal/mol	SD of dG(H2O)_mean
18857	YTYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYD...	M01Y	NaN	0.680	kcal/mol	ddG(mAvg)_mean
18858	YTYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYD...	M01Y	NaN	2.691	M (Molar)	Cm
18859	YTYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYD...	M01Y	NaN	4.519	kcal/mol	dG(H2O)_mean
18860	YTYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYD...	M01Y	NaN	4.630	kcal/mol	dG(mAvg)_mean

18861 rows × 6 columns

Next we do the following: - Select rows that have the value 'ddG(mAvg)_mean' in the 'Assay/Protocol' column. - Keep only the desired columns, and given them shorter names - Remove position 1 from variant sequences and drop duplicate sequences - Flip the sign of measured folding energies - Drop variants with \(\Delta G\) values of exactly \(+4\) kcal/mol, as these were not precisely measured. - Save the dataframe if desired

# Select rows that have the value `'ddG(mAvg)_mean'` in the `'Assay/Protocol'` column.
data_df = raw_df[raw_df['Assay/Protocol']=='ddG(mAvg)_mean'].copy()

# Keep only the desired columns, and given them shorter names
data_df.rename(columns={'Sequence':'x', 'Data': 'y', 'Description':'name'}, inplace=True)
cols_to_keep = ['x', 'name', 'y']
data_df = data_df[cols_to_keep]

# Remove position 1 from variant sequences and drop duplicate sequences
data_df['x'] = data_df['x'].str[1:]
data_df.drop_duplicates(subset='x', keep=False, inplace=True)

# Flip the sign of measured folding energies
data_df['y'] = -data_df['y']

# Drop variants with $\Delta G$ of exactly $+4$ kcal/mol, as these were not precisely measured.
ix = data_df['y']==4
data_df = data_df[~ix]
data_df.reset_index(inplace=True, drop=True)

# Save to file (uncomment to execute)
# data_df.to_csv('nisthal_data.csv.gz', index=False, compression='gzip')
data_df

	x	name	y
0	AYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	T02A	0.4704
1	DYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	T02D	0.5538
2	EYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	T02E	-0.1299
3	FYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	T02F	-0.3008
4	GYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	T02G	0.6680
...	...	...	...
808	TYTLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	K04T	-0.4815
809	TYVLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	K04V	0.2696
810	TYYLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	K04Y	-0.8246
811	VYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	T02V	-1.3090
812	YYKLILNGKTLKGETTTEAVDAATAEKVFKQYANDNGVDGEWTYDD...	T02Y	-0.1476

813 rows × 3 columns