metadata
configs:
- config_name: full
default: true
data_files:
- split: validation
path: valid_full.parquet
- config_name: 10k
data_files:
- split: validation
path: valid_10k.parquet
Evolutionary constraint prediction
Reproducing this dataset
import pandas as pd
CENTER_POSITION = 255
NUCLEOTIDES = list("ACGT")
df = pd.read_csv("hf://datasets/kuleshov-group/cross-species-single-nucleotide-annotation/Evolutionary_constraint/valid.tsv", sep="\t")
df = df.rename(columns={"sequences": "seq"})
df["chrom_pos"] = df.pos
df["pos"] = CENTER_POSITION
df["ref"] = df.seq.str[CENTER_POSITION]
def subsample(df, n, seed):
return (
df.groupby('label').apply(lambda x: x.sample(n=n // 2, random_state=seed))
.reset_index(drop=True)
.sort_values(["chrom", "chrom_pos"])
)
df.drop(columns=["ref"]).to_parquet("valid_full.parquet", index=False)
subsample(df[df.ref.isin(NUCLEOTIDES)].drop(columns=["ref"]), 10_000, 42).to_parquet("valid_10k.parquet", index=False)