Skip to content

Apheris Foundry🔗

Drug discovery workflows in your environment🔗

Apheris Foundry is our product for running, customizing and embedding drug discovery models inside your own infrastructure, built so your agents can use it from day one. Teams can predict molecular structures, benchmark predictions against experimental structures, fine-tune models on their own data, and work with binding-affinity and ADMET capabilities.

Foundry connects these capabilities into reusable workflows. For example, the structure fine-tuning workflow prepares training and validation structures, obtains sequence alignments, builds a training dataset and fine-tunes co-folding models (such as OpenFold3). It returns model checkpoints, training and validation metrics, giving scientists the material to assess how the adapted model performs on their validation data before applying it to their drug discovery programmes.

Tasks can also be incorporated into an existing scientific pipeline. This lets teams reuse individual capabilities while retaining their own analysis tools and scientific decisions.

How the building blocks fit together🔗

---
config:
  flowchart:
    curve: stepAfter
  themeVariables:
    fontFamily: Inter, Arial, Helvetica, sans-serif
---
flowchart LR
    M["Models<br/><br/>Open-source, partner<br/>or network-trained<br/>models"]
    T["Tasks<br/><br/>Prepare data, predict,<br/>fine-tune or<br/>calculate metrics"]
    W["Workflows<br/><br/>Connect tasks into<br/>a scientific process"]
    H["Foundry Hub<br/><br/>Discover workflows,<br/>start runs and<br/>access results"]
    D["Direct task execution<br/><br/>Use tasks in your own<br/>environment or pipeline"]
    M -->|Used by<br/>scientific tasks| T
    T -->|Combined<br/>into| W
    W -->|Delivered<br/>through| H
    T -->|Also available<br/>through| D
  • Models provide learned capabilities, such as predicting a molecular structure.
  • Tasks perform defined operations with specified inputs and outputs. Some tasks, including data preparation, work independently of a model.
  • Workflows connect tasks to complete a scientific process from inputs to results, such as adapting a structure model to a team's data. They manage dependencies between tasks and collect the results.
  • Foundry Hub gives users and their tools access to workflows, execution status and results in the deployed environment.

Foundry is designed for scientists, computational teams and AI agents. Command-line access and workflow instructions allow existing tools and agents to use the same scientific capabilities. Selected workflows also provide guided browser interactions through Foundry Hub.

A catalogue for your team's work🔗

Foundry offers a catalogue of tasks and workflows that can be licensed for your organisation. Our current catalogue groups capabilities into four categories:

  • Predict: molecular structures (including co-folding), binding affinity and ADMET.
  • Benchmark: structure predictions against experimental structures.
  • Fine-tune: structure, binding-affinity and ADMET models.
  • Data preparation: molecular structure datasets.

Your licence and supplied release define the models and capabilities included in your deployment.

Currently co-folding scope includes OpenFold3 and Protenix inference, with structural fine-tuning for OpenFold3. ADMET prediction and fine-tuning workflows are available through Foundry Hub via the CLI.

Foundry and federated learning🔗

Apheris federated networks train models across participating organisations while each organisation's raw data stays in its own environment. When a network-trained model is made available to your organisation, Foundry can bring it into your own prediction, evaluation and customisation workflows.

Foundry can also be used with open-source and partner models. Joining an Apheris federated network is not required to use Foundry. The Apheris Gateway provides the controlled computing environment for federated work.

Start here🔗

Read Tasks and Workflows for examples of scientific inputs and results, then Getting Started with Foundry for customer deployment and first-run guidance. Detailed installation and task guides are available from your Apheris representative or support@apheris.com.