Welcome to Apheris

For training better AI models for life sciences, a larger and more diverse data space is needed. Such data often resides beyond organizational boundaries, and its protection is of utmost importance.

The Apheris Gateway is a federated end-to-end solution for building and joining data networks. The Gateway is designed and developed as a security-hardened, privacy-preserving, and scalable product, providing multi-layered safeguards to protect data confidentiality and IP within data networks.

Capabilities the Apheris Gateway Delivers

• Data Residency: Data always stays within the environment of the Data Custodian.

• Computational Governance: Data Custodians always fully control their data and control computations that visit it via Computational Governance.

• Secure-by-design: A productized, multi-layered approach to protect data, IP, and the confidentiality of every participant within the network.

• Scalable and compatible: The Gateway supports many concurrent workloads across distributed datasets. It works with any off-the-shelf model, such as OpenFold 3, XGBoost, and NVIDIA's BioNeMo library, or custom models and third-party tooling, via a programmatic interface.

Computational Governance is the core concept that makes privacy-preserving data networks possible, which gives Data Custodians control over who can run which model on which dataset and under what conditions. Computational Governance is highly adaptive and can enable a wide range of federated model training tasks within life sciences.

Stay in Full Control: Computational Governance

Computational Governance is a method for controlling, supervising, and tracking all aspects of computations. It works by automatically evaluating incoming compute requests, enforcing privacy and security parameters, and overseeing the release of results. This allows Data Custodians to contribute any data type safely to collaborations while preserving data privacy.

In a nutshell, Computational Governance within Apheris allows control of any Docker-based application on the algorithmic level.

Computational Governance enables Data Custodians to enforce security, confidentiality, and auditability at the computation level.

• Security: Only approved users can send computations to the Gateway. All network communication is encrypted following industry standards like TLS 1.2+.

• Confidentiality: Data privacy protection during collaborations is a function of data, model, model permissions, and privacy-enhancing technologies. Computational Governance allows for finding the optimal setting for both training success and data protection.

• Auditability: The Gateway tracks all interactions with data registered to it and provides an interface to persist these audit logs to your existing tooling.

Asset Policies are rules set by the Data Custodian that define the allowed operations by named users on specified datasets. Read the Asset Policy page for more details.

Compute Specifications (Compute Spec) are defined by the data scientist, who specifies what computation they intend to run on which data. Read the Compute Spec page for more details.

Asset Policies and Compute Specs form a computational agreement, ensuring safe data use. Only computations meeting the requirements outlined in an Asset Policy can be executed, and detailed logs for each data interaction of computations are created for the Data Custodian to demonstrate compliance.

For Data Custodians

After deploying the Apheris Gateway, custodians use the Governance Portal to:

• Register datasets

• Define and manage Asset Policies

• Approve computations

• Manage user permissions

Security teams can review the architecture. Further details on compliance, controls, and penetration test results can be accessed in the Apheris Trust Center.

For Data Scientists

Apheris fits seamlessly into existing ML workflows via a CLI. With it, data scientists can:

• Explore dataset metadata and policy requirements

• Submit Compute Specs

• Use registered models or custom models

• Run simulations and monitor compute jobs

No need to change tools or platforms.

A Federated Workflow

Apheris enables a simple, structured collaboration process.

• Custodians define Asset Policies for registered datasets.

• Data scientists submit compliant Compute Specs.

• The Apheris Gateway validates, automatically provisions, and executes Compute Specs within the Data Custodians environment. The aggregation of results happens securely on the Orchestrator.

Sensitive data never leaves the Data Custodian's environment.