SEA26

From DISI
Revision as of 00:34, 5 November 2026 by Khtang (talk | contribs)
Jump to navigation Jump to search

This is the documentation for the new version of SEA by Yujin Wu and Khanh Tang.

SEA26
Sea homepage.png
Paragigm Docking
Developer Shoichet Lab
Stable Version SEA
Programming Language Python,JS
Dependencies Boltz2,ReactJS,FastAPI

About

The Similarity Ensemble Approach (SEA) web server predicts which protein targets a small molecule is likely to bind, based on how chemically similar it is to known ligands of those targets. Submit one or more compounds as SMILES strings and SEA returns a ranked list of targets sorted by statistical significance.

Citing SEA

Please cite these publications below:

  • Keiser MJ, Roth BL, Armbruster BN, Ernsberger P, Irwin JJ, Shoichet BK. Relating protein pharmacology by ligand chemistry. Nat Biotech 25 (2), 197-206 (2007)[1].
  • TBD

Submitting a SEA job

GUI

Paste one compound per line in the input box on the home page. Each line can be a SMILES string alone, or a SMILES string followed by a compound name:

COc1ccc2[nH]cc(c2c1)C(CC)CN
Cc1ccc2c3c1C(CN3CCNC2)CC aspirin_analog
  • Maximum: 10 compounds per submission
  • Compound names are optional; if omitted they are auto-labeled compound_1, compound_2, etc.

Click ```Submit```

API

1. Authentication

Log in to get a token:

curl -s -X POST https://sea.docking.org/api/auth/login \
 -H "Content-Type: application/json" \
 -d '{"email": "you@example.com", "password": "yourpassword"}'

Response:

{"access_token": "<jwt_token>", "first_name": "...", "detail": "Login successfully!"}

2. Submit a Job

curl -s -X POST https://sea.docking.org/api/submit_sea \
 -H "Authorization: Bearer <access_token>" \
 -H "Content-Type: application/json" \
 -d '{"query": "COc1ccc2[nH]cc(c2c1)C(CC)CN my_compound"}'

Response:

{"task_id": "<task_id>"}

Error responses:

HTTP Status Meaning
400 Empty query, >10 compounds, or invalid SMILES
401 Missing or expired Bearer token
429 Server is busy — retry later

3. Poll for Results

curl -s "https://sea.docking.org/api/result?taskId=<access_token>"

4. Download Results

curl -s -O -J "https://sea.docking.org/api/download?taskId=<access_token>"

Returns a ZIP containing `sea_result.tsv` and per-target `top10_<chembl_id>.tsv` files.

Result explained

Results Table Columns

alt text

Main Columns

Column Description
Target ChEMBL ID Protein ChEMBL ID
Target Name Human-readable protein name
UniProt ID / Accession UniProt identifiers
Z-score How many standard deviations above the mean the similarity score is
p-value Statistical significance; smaller = more significant
MaxTc Highest Tanimoto coefficient between any query compound and a known ligand of this target (0–1 scale)

Boltz2 Affinity Columns

After SEA completes, the server automatically queues structure predictions for the top 5 novel targets (single-compound submissions only).

Column Description
Affinity (kcal/mol) Predicted binding free energy (ΔG); more negative = tighter predicted binding
Binding probability Model confidence that the compound binds (0–100%)

Navigating Compound vs Targets Heatmap

The Compound vs Targets view gives a full-matrix overview of how your query compound relates to every target returned by SEA. Click the Compound VS Targets button at the top of the results page to open it.

Layout

Panel Contents
Left — heatmap grid Rows = targets; columns = known ChEMBL ligands for each target; cell color = Tanimoto similarity score between your query compound and that ligand
Right — detail panel Query compound structure (always visible); known hit structure and metadata on hover or click

Cell colors run from red/orange (low similarity) to dark green (Tc ≈ 1.0).

Navigating the Heatmap

alt text
  1. Hover over any cell to preview the known hit structure and its Tanimoto score in the right panel
  2. Click a cell to lock (pin) the selection — the right panel stays fixed so you can read the metadata without holding the mouse
  3. Use the arrow keys to move between cells while a cell is locked
  4. Press Enter or click the same cell again to unpin

Reading the Detail Panel

When a cell is selected the right panel shows:

  • Query compound — your submitted structure (always shown)
  • Known hit — the ChEMBL ligand in that cell
  • Target name and ChEMBL ID
  • UniProt accession
  • Tanimoto score for the query–hit pair

Submitting a Boltz2 Job from the Heatmap

User must log in in order to use this feature

With a cell locked, two submission buttons appear in the right panel:

Button What it predicts
Submit Query × Target 3D complex of your query compound docked to the selected target
Submit Both × Target Both the query compound and the known hit docked to the target, for direct comparison

After submission the button is replaced by a status indicator (queued → running → done). Click Download when the job finishes to get the structure files and a ChimeraX visualization script.


FAQ

TBD

Data sources

TBD

Software is open source

Github Repo

How to do an alignment

  • co-fold the query ligand and separately a similar ligand
  • download both to your computer (takes 10 mins +)
  • use chimeraX (UCSF ChimeraX) to load the CIF file of each
  • use matchmaker #2 #1 to align the proteins
  • use select ligand to focus on the ligand.
  • visually inspect the query ligand and the most similar annotated ligand.
  • do they occupy the same space?
  • do they interact with the same residues?
  • do the small molecules align in the part of the molecule that is recognized by the protein?
  • evaluate whether it seems likely that this prediction is correct? that your query ligand would be expected to recapitulate the same interactions as a known ligand.
  • check also for any available experimental structures in the PDB and consider aligning them.