Specifying Target Organisms in Nvidia NIM API for Evo2 Sequence Generation

[MarcAmil30]

In the Evo Designer, there is an option to select the organism for which you want to generate sequences (e.g., humans, E. coli, baker's yeast) before inputting your sequence.

I would like to know how this functionality is implemented in the Nvidia NIM API version or when installing it, as there seems to be no option to specify the target organism in both examples below:

from evo2 import Evo2

evo2_model = Evo2('evo2_7b')

output = evo2_model.generate(prompt_seqs=["ACGT"], n_tokens=400, temperature=1.0, top_k=4)

print(output.sequences[0])

#!/usr/bin/env python3
import requests
import os
import json
from pathlib import Path

key = os.getenv("NVCF_RUN_KEY") or input("Paste the Run Key: ")

r = requests.post(
    url=os.getenv("URL", "https://health.api.nvidia.com/v1/biology/arc/evo2-40b/generate"),
    headers={"Authorization": f"Bearer {key}"},
    json={
        "sequence": "ACTGACTGACTGACTG",
        "num_tokens": 8,
        "top_k": 1,
        "enable_sampled_probs": True,
    },
)

if "application/json" in r.headers.get("Content-Type", ""):
    print(r, "Saving to output.json:\n", r.text[:200], "...")
    Path("output.json").write_text(r.text)
elif "application/zip" in r.headers.get("Content-Type", ""):
    print(r, "Saving large response to data.zip")
    Path("data.zip").write_bytes(r.content)
else:
    print(r, r.headers, r.content)

[garykbrixi]

Hi Marc. We use a util function to generate the phylogenetic tags which is currently not included in the Nvidia NIM. The phylogenetic tags can be generated easily using taxonomic information.

I've added our util function to do this here and it will be merged to main after review

[MarcAmil30]

Thank you very much!

Just to clarify once the util function is merged to main this functionality will still not be included in the Nvidia NIM but I can use it when I install Evo through method as shown below (using pip install .) ? Lastly did you find an improvement of performance of the generation of species-specific sequence when using the phylogenetic tags compared to relying on semantic mining where you used it on Evo on a previous publication.

git clone --recurse-submodules git@github.com:ArcInstitute/evo2.git
cd evo2
pip install .

Thanks again.

[garykbrixi]

Yes that's right. You can start using the tag generation function right now if you copy it

We did not do a systematic comparison of this. The tags combined with prompts seemed to benefit on long sequence generations compared to just DNA prompts.

Note that only the 1M models can use phylo tags since the tags are added in midtraining, they should not be used with the base models which did not see them in pretraining