Maps protein interactions, ligand relationships, and functional annotation context from sequence.

Atlas

Atlas is Synthyra's sequence-first protein intelligence platform. It connects protein-protein interaction prediction, protein-ligand prioritization, and CAMP-based functional annotation so researchers can move from raw sequence to ranked biological hypotheses.

The output is not experimental truth. Atlas is a way to make large search spaces small enough to inspect, test, and act on.

What It Does

Atlas supports three connected analysis lanes:

• PPI maps likely protein-protein relationships from amino acid sequence.
• PLI prioritizes protein-ligand hypotheses for target exploration, repurposing, and early screening.
• CAMP connects sequences to structured functional annotation space.

Those lanes are most useful together. An interaction neighborhood is easier to interpret when functional annotations and ligand hypotheses are visible in the same workflow.

Why It Matters

Biology is combinatorial. A human proteome contains roughly 20,000 proteins, which creates hundreds of millions of possible protein pairs before ligands, pathogen proteins, generated designs, or functional annotations enter the analysis.

Atlas treats that scale as the starting point. Once proteins are embedded, interaction screens can be scored as matrices rather than as isolated pairwise questions. That makes whole-proteome, inter-actome, and query-centered analyses practical enough for discovery workflows.

The research direction behind Atlas is also shaped by hard evaluation lessons: same-species negative controls, cluster-aware splits, and homology checks help reduce shortcuts such as taxonomy leakage and sequence-neighborhood memorization.

Using Atlas

Score matched protein pairs over HTTPS with a Synthyra API key:

import requests

api_key = "..."  # synthyra.com/settings?section=api-keys

resp = requests.post(
    "https://api.synthyra.com/v1/score/pairs",
    headers={"Authorization": f"Bearer {api_key}"},
    json={
        "inputs_a": ["MEV...", "MAK..."],
        "inputs_b": ["MQT...", "MLG..."],
        "ids_a": ["P04637", "P12345"],
        "ids_b": ["P38398", "Q9Y6K9"],
        "both_directions": True,
    },
    timeout=180,
)
result = resp.json()

/v1/score/pairs returns quantized confidence scores in [0, 100]. For all-vs-all matrices use /v1/score/matrix; for network expansion around query proteins use /v1/generate/network.

Intended Use

Use Atlas to rank interaction partners, expand protein neighborhoods, compare engineered proteins against reference proteomes, prioritize ligand hypotheses, and add structured function context to poorly characterized sequences.

Limitations

Atlas should be interpreted as a predictive research system, not a source of ground truth.

• PPI scores do not prove direct physical contact or measure affinity.
• PLI scores do not prove binding, efficacy, selectivity, or safety.
• CAMP suggestions do not prove enzyme activity, pathway participation, localization, or biological role.
• Cellular context, expression, localization, cofactors, post-translational modification, and conformational state can change whether a predicted relationship occurs in vivo.
• Novel designed proteins, unusual ligands, and emerging pathogens can fall outside the strongest parts of the training distribution.

For consequential decisions, combine Atlas with structural modeling, STRING, BioGRID, UniProt, assay data, medicinal chemistry review, domain expertise, and wet-lab validation.