drugclaw/bio-tools

Bio Tools

Use this skill for bioinformatics, genomics, transcriptomics, structural biology, or general biological data-analysis requests.

This skill focuses on DrugClaw's reproducible bioinformatics workflow pattern:

• Run commands with bash
• Write reproducible scripts with write_file or edit_file
• Inspect outputs with read_file
• Use web_search / web_fetch for literature and database lookups

Environment Check

Do not assume the runtime already has the biology stack. Check first.

which python3 blastn blastp samtools bedtools bwa minimap2 fastqc seqtk pymol || true
python3 - <<'PY'
mods = ["Bio", "pandas", "numpy", "matplotlib", "pysam", "seaborn", "sklearn"]
for name in mods:
    try:
        __import__(name)
        print(f"{name}: ok")
    except Exception as exc:
        print(f"{name}: missing ({exc})")
PY

If key tools are missing, say so explicitly and recommend the optional drug-sandbox image documented in docs/operations/science-runtime.md.

Working Style

• Start by inventorying files and formats in the current chat working directory.
• State the exact command, parameters, and tool versions used.
• Save outputs as files in the working directory and mention their paths in the reply.
• Prefer reproducible scripts over one-off interactive commands for multi-step analysis.
• Flag quality issues, contamination risk, reference mismatch, and missing metadata.
• End with concrete next steps, not just raw output.

Common Workflow

1. Inspect available inputs.
2. Identify data type: FASTA, FASTQ, BAM, BED, CSV/TSV, PDB, SMILES, etc.
3. Check tool availability.
4. Run the smallest validating command first.
5. Save primary outputs and a short README or command log when analysis is non-trivial.
6. Summarize findings with caveats and suggested follow-ups.

File Triage

pwd
find . -maxdepth 3 -type f | sort
file sample.fastq.gz
gzip -dc sample.fastq.gz | head

Useful quick checks:

samtools --version | head -n 1
blastn -version
fastqc --version
python3 --version

Sequence Search

# Nucleotide BLAST against a local FASTA database
blastn -query query.fa -subject reference.fa -outfmt 6 -evalue 1e-5 > blast.tsv

# Protein BLAST
blastp -query protein.fa -subject reference_proteins.fa -outfmt 6 > blastp.tsv

# Translate nucleotide query against proteins
blastx -query transcript.fa -subject proteins.fa -outfmt 6 > blastx.tsv

For remote NCBI lookups, prefer Python so the workflow is easy to archive:

from Bio import Entrez, SeqIO

Entrez.email = "[email protected]"
handle = Entrez.efetch(db="nucleotide", id="NM_000546", rettype="fasta", retmode="text")
record = SeqIO.read(handle, "fasta")
print(record.id, len(record.seq))

Read Alignment And BAM Processing

# Build index
bwa index reference.fa

# Short-read alignment
bwa mem reference.fa reads_R1.fastq.gz reads_R2.fastq.gz > aligned.sam

# Long-read alignment
minimap2 -a reference.fa long_reads.fastq.gz > aligned.sam

# SAM -> sorted/indexed BAM
samtools view -bS aligned.sam | samtools sort -o aligned.sorted.bam
samtools index aligned.sorted.bam
samtools flagstat aligned.sorted.bam > aligned.flagstat.txt

FASTQ Quality Control

mkdir -p qc
fastqc reads_R1.fastq.gz reads_R2.fastq.gz -o qc

# Quick sequence statistics
seqtk comp reads_R1.fastq.gz | head
seqtk size reads_R1.fastq.gz

Report at minimum:

• total reads
• adapter or overrepresented sequence warnings
• per-base quality drop-off
• GC bias
• whether trimming/filtering is needed before alignment

Genome Arithmetic

bedtools intersect -a peaks.bed -b genes.bed > overlap.bed
bedtools coverage -a targets.bed -b aligned.sorted.bam > coverage.tsv
bedtools getfasta -fi reference.fa -bed targets.bed > targets.fa

Python Analysis Recipes

Sequence I/O

from Bio import SeqIO

for record in SeqIO.parse("input.fa", "fasta"):
    print(record.id, len(record.seq))

Differential Expression

import pandas as pd
from pydeseq2.dds import DeseqDataSet
from pydeseq2.ds import DeseqStats

counts = pd.read_csv("counts.csv", index_col=0)
meta = pd.read_csv("metadata.csv", index_col=0)

dds = DeseqDataSet(counts=counts, metadata=meta, design="~condition")
dds.deseq2()
stats = DeseqStats(dds, contrast=["condition", "treated", "control"])
stats.summary()
res = stats.results_df.sort_values("padj")
res.to_csv("deseq2_results.csv")

Single-Cell RNA-seq

import scanpy as sc

adata = sc.read_h5ad("data.h5ad")
sc.pp.normalize_total(adata)
sc.pp.log1p(adata)
sc.pp.highly_variable_genes(adata)
sc.pp.pca(adata)
sc.pp.neighbors(adata)
sc.tl.umap(adata)
sc.tl.leiden(adata)
sc.pl.umap(adata, color="leiden", save="_leiden.png")

Publication-Style Plots

import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns

df = pd.read_csv("expression.csv")
sns.set_theme(style="whitegrid")
ax = sns.scatterplot(data=df, x="log2FoldChange", y="-log10_padj", hue="significant")
ax.figure.savefig("volcano.png", dpi=300, bbox_inches="tight")

Structural Biology

Fetch structures from PDB with web_fetch or direct download, then render with PyMOL if available.

curl -L https://files.rcsb.org/download/1M17.pdb -o 1M17.pdb
cat > render.pml <<'PML'
load 1M17.pdb, prot
hide everything
show cartoon, prot
spectrum count, rainbow, prot
bg_color white
png 1M17_rainbow.png, width=1600, height=1200, dpi=200, ray=1
quit
PML
pymol -cq render.pml

When PyMOL is unavailable, still provide the fetched structure, any residue/chain findings, and the exact rendering script the user can run later.

Literature Search

Use web_search or PubMed APIs for recent papers. For structured PubMed workflows:

from Bio import Entrez

Entrez.email = "[email protected]"
search = Entrez.esearch(db="pubmed", term="CRISPR off-target 2025[dp]", retmax=5)
ids = Entrez.read(search)["IdList"]
summary = Entrez.esummary(db="pubmed", id=",".join(ids))
print(Entrez.read(summary))

Summaries should include:

• citation
• study type
• model system
• main finding
• why it matters for the user's question

Outputs

Good replies should mention:

• what inputs were used
• what commands/scripts were run
• what files were generated
• the most important biological conclusion
• uncertainty or validation limits

Example closing pattern:

I aligned the reads against GRCh38 with bwa mem and generated `aligned.sorted.bam` plus `aligned.flagstat.txt`.
FastQC shows 3' quality decay after cycle 125 and adapter contamination in R2, so trimming before re-alignment is recommended.
Next step: run fastp or cutadapt, then repeat alignment and variant calling.

Related Skill

For remote biology database lookups across UniProt, PDB, AlphaFold, ClinVar, Ensembl, GEO, InterPro, KEGG, OpenTargets, Reactome, or STRING, activate bio-db-tools. For AnnData, single-cell dataset profiling, alignment-region inspection, or mzML inventory, activate omics-tools. For public drug-discovery database lookups across PubChem, ChEMBL, openFDA, ClinicalTrials.gov, or OpenAlex, activate pharma-db-tools. For molecular docking or pose inspection, activate docking-tools. For DeepChem, PySCF, RDKit descriptors, or chemistry-specific follow-up, activate chem-tools.