Plasmid Optimizer

Optimize plasmid sequences (DNA or amino acid) for bacterial expression. Choose constraints (codon usage, GC content, restriction sites, secondary structure, repeats/homopolymers) and generate novel peptides with PepMLM (tunable novelty: top-k, temperature). Delivered as a web app and CLI.

Install

From the project root:

pip install -e .

For novel peptide generation (PepMLM):

pip install -e ".[ml]"

Optional: if you hit NumPy/pandas compatibility issues with DnaChisel, try numpy<2 in your environment.

Run the API and Web UI

From the project root (with the package installed):

uvicorn api.main:app --reload --host 0.0.0.0 --port 8000

Open http://localhost:8000 for the web UI. The same server serves the REST API.

UniProt integration

You can load a protein sequence directly from UniProt by accession or FASTA URL.

• Web UI: In "Generate peptide binder" or "Sequence to optimize", enter a UniProt accession (e.g. Q6JKW3) or paste a FASTA URL (e.g. https://rest.uniprot.org/uniprotkb/Q6JKW3.fasta) and click Fetch to fill the sequence.
• CLI: Use --uniprot-id Q6JKW3 (or a full FASTA URL) instead of -s/-i; the sequence is fetched and treated as amino acid.

CLI

# Optimize an amino acid sequence (output to stdout)
plasmid-optimize -s "MKQL" -t aa

# Fetch from UniProt and optimize
plasmid-optimize --uniprot-id Q6JKW3 -o optimized_dna.txt

# Optimize from file, write DNA to file
plasmid-optimize -i sequence.txt -t aa -o optimized_dna.txt

# With constraints
plasmid-optimize -s "MKQL..." -t aa --organism e_coli --gc-min 0.4 --gc-max 0.6 --avoid-enzymes EcoRI,BamHI

# Generate novel peptides for a target protein (PepMLM; tune novelty with top-k and temperature)
plasmid-optimize --generate-binder-for-target "MKTIIALSYIFCL..." --peptide-length 15 --num-binders 4 --top-k 3 --temperature 1.0

# JSON output
plasmid-optimize -s "MKQL" -t aa --json -o result.json

Constraints

Constraint	Description
Codon usage	Organism-specific (e.g. e_coli, s_cerevisiae).
GC content	Min/max and sliding window size.
Restriction sites	Avoid selected enzymes (e.g. EcoRI, BamHI, BsaI).
Secondary structure	Avoid hairpins (DnaChisel AvoidHairpins).
Repeats / homopolymers	Uniquify kmers; avoid long homopolymer runs.

API

• POST /optimize — Body: { "sequence", "sequence_type": "aa"|"dna", "constraints": {...} }. Returns optimized_dna, amino_acid, report.
• POST /generate-binder — Novel peptide generation. Body: { "target_protein_sequence", "peptide_length", "num_binders", "top_k", "temperature" }. Returns peptides. Requires [ml]. Higher top_k and temperature increase diversity/novelty.
• GET /species — List supported organisms for codon optimization.
• GET /fetch-uniprot?id=Q6JKW3 — Fetch protein sequence from UniProt by accession or FASTA URL. Returns { "sequence", "header" }.

Novel peptide generation (PepMLM)

Parameter	Description
peptide_length	Length of generated peptide (masked positions).
num_binders	Number of novel sequences to generate.
top_k	Top-k candidates per position; higher = more diversity.
temperature	Sampling temperature; higher = more novel/random.

Hosting with PepMLM (GCP)

To run the app with PepMLM (e.g. on Google Cloud), store the model in GCS and run the API with PEPMLM_MODEL_PATH or PEPMLM_GCS_URI. See HOSTING.md for step-by-step GCP setup (save model to GCS, Cloud Run, or GCE VM).

References

• DnaChisel — sequence optimization.
• ViennaRNA — RNA structure (optional).
• PepMLM (ChatterjeeLab/PepMLM-650M) — novel peptide generation.