CLI Reference

Molfun provides a command-line interface built with Typer. The entry point is molfun.

molfun --help

Commands Overview

Command	Description
molfun structure	Predict protein structure from sequence
molfun affinity	Predict protein-ligand binding affinity
molfun info	Display model and environment information
molfun fetch-pdb	Download PDB structures
molfun fetch-msa	Generate/fetch MSAs for sequences
molfun fetch-domain	Fetch domain annotations
molfun parse	Parse structure/sequence files
molfun registry	List registered modules
molfun agent	Run an AI agent for automated workflows
molfun eval	Evaluate a model on a benchmark
molfun benchmark	Run performance benchmarks
molfun run	Run a training/export script
molfun push	Push a model to the Hub
molfun pull	Pull a model from the Hub
molfun push-dataset	Push a dataset to the Hub

---

structure

Predict the 3D structure of a protein.

molfun structure \
    --sequence "MKFLILLFNILCLFPVLAADNH..." \
    --model openfold_v2 \
    --output prediction.pdb \
    --device cuda \
    --num-recycles 3

Flag	Type	Default	Description
--sequence / -s	str	required	Amino acid sequence
--fasta / -f	Path	--	FASTA file with sequence(s)
--model / -m	str	openfold_v2	Pretrained model name
--output / -o	Path	prediction.pdb	Output PDB file
--device	str	cpu	Compute device
--num-recycles	int	3	Recycling iterations
--msa	Path	--	Precomputed A3M MSA file

---

affinity

Predict binding affinity for a protein-ligand pair.

molfun affinity \
    --sequence "MKFLILLFNILCLFPVLAADNH..." \
    --ligand ligand.sdf \
    --model openfold_v2

Flag	Type	Default	Description
--sequence / -s	str	required	Protein sequence
--ligand / -l	Path	required	Ligand file (SDF)
--model / -m	str	openfold_v2	Model name
--device	str	cpu	Compute device

---

info

Display information about the Molfun installation and available models.

molfun info
molfun info --model openfold_v2

Flag	Type	Description
--model / -m	str	Show details for a specific model
--verbose / -v	bool	Show extended system information

---

fetch-pdb

Download PDB structures from RCSB.

# By IDs
molfun fetch-pdb --ids 1ABC 2DEF 3GHI --output-dir ./pdb_files

# By query
molfun fetch-pdb \
    --resolution-max 2.5 \
    --organism "Homo sapiens" \
    --max-results 100 \
    --output-dir ./pdb_files

Flag	Type	Default	Description
--ids	list[str]	--	PDB IDs to download
--resolution-max	float	--	Max resolution filter
--organism	str	--	Source organism filter
--method	str	--	Experimental method filter
--max-results	int	--	Maximum entries to download
--output-dir / -o	Path	./pdb	Output directory
--format	str	mmcif	File format (pdb or mmcif)

---

fetch-msa

Generate or fetch MSAs for protein sequences.

molfun fetch-msa \
    --sequence "MKFLILLFNILCLFPVLAADNH..." \
    --output alignment.a3m \
    --backend colabfold

Flag	Type	Default	Description
--sequence / -s	str	--	Amino acid sequence
--fasta / -f	Path	--	FASTA file with sequence(s)
--output / -o	Path	alignment.a3m	Output A3M file
--backend	str	colabfold	MSA backend

---

fetch-domain

Fetch domain annotations for a protein.

molfun fetch-domain --uniprot P12345

Flag	Type	Description
--uniprot / -u	str	UniProt accession
--pdb	str	PDB ID

---

parse

Parse structure and sequence files.

molfun parse protein.pdb --format pdb
molfun parse alignment.a3m --format a3m --stats

Flag	Type	Default	Description
FILE	Path	required	Input file (positional)
--format	str	auto	File format (auto-detected from extension)
--stats	bool	False	Print summary statistics
--output / -o	Path	--	Output file for converted data

---

registry

List registered modular components.

molfun registry                   # list all registries
molfun registry attention         # list attention modules
molfun registry --details block   # detailed info per block

Flag	Type	Description
TYPE	str	Optional: registry type to list (attention, block, embedder, structure_module, loss)
--details / -d	bool	Show detailed information for each entry

---

agent

Run an AI agent for automated experiment workflows.

molfun agent "Fine-tune openfold on CASP15 with LoRA rank 8"

Flag	Type	Default	Description
PROMPT	str	required	Natural language instruction (positional)
--model	str	gpt-4	LLM model to use
--dry-run	bool	False	Print plan without executing

---

eval

Evaluate a model on a benchmark dataset.

molfun eval \
    --model openfold_v2 \
    --benchmark casp15 \
    --device cuda \
    --output results.json

Flag	Type	Default	Description
--model / -m	str	required	Model name or path
--benchmark / -b	str	required	Benchmark name (e.g., casp15, cameo)
--device	str	cpu	Compute device
--output / -o	Path	results.json	Output file

---

benchmark

Run performance benchmarks (throughput, latency, memory).

molfun benchmark \
    --model openfold_v2 \
    --sequence-lengths 100 200 500 \
    --device cuda \
    --warmup 5 \
    --iterations 20

Flag	Type	Default	Description
--model / -m	str	required	Model name
--sequence-lengths	list[int]	[100, 200, 500]	Sequence lengths to benchmark
--device	str	cpu	Compute device
--warmup	int	5	Warmup iterations
--iterations	int	20	Benchmark iterations

---

run

Run a training or export script.

molfun run train.py --gpus 4
molfun run export --format onnx --model openfold_v2 --output model.onnx

Flag	Type	Default	Description
SCRIPT	str	required	Script path or subcommand
--gpus	int	1	Number of GPUs (launches with torchrun for > 1)

---

push / pull

Push and pull models from the Molfun Hub.

# Push
molfun push ./my_model --repo myorg/finetuned-openfold --private

# Pull
molfun pull myorg/finetuned-openfold --output ./my_model

push

Flag	Type	Default	Description
PATH	Path	required	Local model directory
--repo / -r	str	required	Hub repository ID
--private	bool	False	Make repository private
--message	str	--	Commit message

pull

Flag	Type	Default	Description
REPO	str	required	Hub repository ID
--output / -o	Path	./	Output directory
--revision	str	--	Specific revision/tag

---

push-dataset

Push a dataset to the Hub.

molfun push-dataset ./my_dataset --repo myorg/structure-dataset --format parquet

Flag	Type	Default	Description
PATH	Path	required	Local dataset directory
--repo / -r	str	required	Hub repository ID
--format	str	parquet	Dataset format
--private	bool	False	Make repository private