MARCA 369

Cookies

Only with your permission do we use Microsoft Clarity to measure, in aggregate, how the site is used. We load no marketing scripts. The choice is stored on this device.

Customise
Customise

Necessary, always on: the bag session and this choice.

A molecule turning into a data stream feeding prediction dashboards

Technology · In-house ADMET model

Our molecular model

An AI model built by Marca 369 for three ADMET tasks from molecular structure. Runs on CPU, no GPU, evaluated on the public TDC benchmark (5-seed protocol).

For research only, not for clinical decisions.

TDC · Therapeutics Data Commons

Results on TDC

Mean ± std (5 seeds) of our model vs. the public TDC leaderboard #1. Only these 3 tasks are offered in the demo and API.

  • 3

    Microsomal clearance

    Best published result on TDC for microsomal clearance

    0.643 ± 0.020 vs 0.630 ± 0.010

  • 6

    Half-life

    On par with 1st place on TDC for half-life

    0.579 ± 0.026 vs 0.576 ± 0.025

  • 9

    CYP3A4 substrate

    On par with 1st place on TDC for CYP3A4 substrate

    0.670 ± 0.025 vs 0.667 ± 0.019

parameters
≈2.8 M
per molecule, on CPU
0.1–0.2 s
trained on benchmark data only
No pre-training

For research only, not for clinical decisions.

SMILES → ADMET

How it works

  • 01

    Input

    The molecule's structure, in SMILES notation.

  • 02

    Representation

    The model reads the molecule as a graph of atoms and bonds.

  • 03

    Prediction

    It returns microsomal clearance, half-life and CYP3A4 substrate probability.

  • 04

    Efficiency

    ≈2.8 M parameters, no pre-training, 0.1–0.2 s per molecule on CPU.

Partnerships

Who it is for

  • 01

    Computational chemistry

    Fast screening of molecule series before the lab.

  • 02

    Teaching & research

    Demonstrations of machine learning applied to chemistry.

  • 03

    Integration

    Partners who want the model on compact hardware or their own platforms.

Hugging Face · benfica306/marca369-bio

Free public demo

Paste a molecule in SMILES and see predictions in seconds. Customers and members get extra free predictions.

Every Marca 369 customer and member gets extra free predictions from our model.

Open the demo ↗Request predictions

For research only, not for clinical decisions.

By request

Equipment on request

Computing, AI, travel, and leisure. Each page names the supplier and the reference price. None of this is bought here.

Tecnologia · Modelo molecular · Marca 369