Apheris Secures $8.25M to Revolutionize AI Data Access in Life Sciences

Apheris rethinks the AI data bottleneck in life science with federated computing | TechCrunch

AI is fundamentally dependent on data, but the vast majority of health data goes unused for understandable reasons -- chiefly patient privacy, regulation and IP protection. "This is the core underlying problem" of building AI solutions for life sciences and related areas like pharmaceutics, said German entrepreneur Robin Röhm. And not only that: collaboration when it comes to sensitive data can be a challenge. Apheris, Röhm's startup, aims to address this through federated computing: making data securely accessible for AI model training without moving it by taking a decentralized approach. Its customers include Roche and several hospitals, he said. The core philosophy of federated computing is that "computations are executed locally where data resides, and only the outputs (e.g., model parameters) are aggregated centrally," says Marcin Hejka, a co-founder and managing partner at OTB Ventures. Hejka has now co-led an $8.25 million Series A into Apheris alongside fellow deep tech investor eCAPITAL. Hejka believes Apheris could become a critical component in the federated data networks that are starting to emerge. "We see a maturing ecosystem of third-party software tools (open-source federation engines, data quality tools, and security products)," he told TechCrunch. "Apheris also enables seamless integration with complementary privacy-enhancing technologies (homomorphic encryption, differential privacy, synthetic data)." Apheris's new funding comes in the wake of a pivot. Originally, Röhm and his co-founder Michael Höh started the company in 2019 with the goal of building a federated learning framework that competed with open source approaches, based on their experiences at their previous startup, Janus Genomics. But after raising a large seed round in 2022, the duo made a major pivot in 2023 to focus on the data owner side and double down on pharma and life sciences. According to Röhm, this paid off. The startup found product-market fit with the new product it launched in the last quarter of 2023, and multiplied its revenue by 4 since then. Also backed by existing investors including Octopus Ventures and Heal Capital, its new round brings its total funding to $20.8 million, which will help the company hire senior talent with life science backgrounds, also on the commercial side. The Apheris Compute Gateway, the software agent that serves as a gateway between local data and AI models, is already being used by the AI Structural Biology (AISB) Consortium, a joint initiative that sees members such as AbbVie, Boehringer Ingelheim, Johnson & Johnson and Sanofi collaborate on AI-driven drug discovery. Protein complex prediction will be one topic Apheris will further focus on with this new funding. While use-case agnostic, it understands that it can add value when there is very limited public data available, yet much more valuable and diverse data that won't be unlocked unless life sciences companies feel safe doing so. "Without addressing the data owners' concerns in providing data to AI, we don't think that the impact of AI can really be unlocked, and that's ultimately the core mission of what we're building," Röhm said.

Apheris raises $8.25M for its healthcare-focused federated AI platform - SiliconANGLE

Apheris raises $8.25M for its healthcare-focused federated AI platform Apheris AI GmbH, a startup with a platform for analyzing healthcare data, today disclosed that it has raised $8.25 million in funding. OTB Ventures and eCAPITAL led the Series A round. The investment brings Berlin-based Apheris' total outside funding to $20.8 million. In the healthcare sector, researchers from different organizations often require the ability to share clinical data with one another. But sending data over the network to a different organization can present cybersecurity risks. Healthcare institutions address those risks using an approach called federated computing, which forms the basis of Apheris' platform. With federated computing, a company doesn't have to transfer its clinical datasets outside the corporate network to make them accessible for third-party researchers. The technology lets researchers remotely run analyses on the part of the company's internal infrastructure that hosts the data. The processing results are subsequently sent back to the researchers over the network without moving any of the information that was analyzed. Apheris' platform makes it easier for healthcare institutions to implement federal computing. It also enables researchers to run artificial intelligence models on the clinical datasets shared with them. Apheris' platform facilitates federated computing projects with a lightweight program, or agent, called the Compute Gateway. It runs on the system that hosts the clinical dataset a company plans to make available for external researchers. Once installed, the Compute Gateway allows researchers to request access to the dataset on a self-service basis. Apheris users structure their information access requests as a so-called Compute Spec. This is a file that specifies what dataset researchers wish to access, the AI model that they plan to run on the dataset and the hardware resources required for the task. Employees at the organization that owns the dataset can use a centralized dashboard to review and approve such requests. Under the hood, the platform is based on an open-source Nvidia Corp. framework called FLARE. When a researcher runs an AI model on a system that hosts a dataset, FLARE ensures the system's operator can't reverse-engineer the model to obtain its training dataset. The framework likewise blocks attempts to access the neural network's weights. According to Apheris, its platform protects not only AI models but also the clinical information they process. Data usually has to be decrypted before it can be analyzed. While in plaintext form, information is considerably easier for hackers to steal, which increases the risk of breaches. Apheris implements a technique called homomorphic encryption that makes it possible to run queries on a dataset without decrypting it. The feature is complemented by a second security feature called differential privacy. When the clinical dataset used in a research project contains personal information such as patient records, scientists often anonymize it before sharing it with third parties. Differential privacy makes it more impossible for hackers to extract personal information from anonymized datasets. Apheris says that its software has been adopted by Johnson & Johnson, Roche and other major pharmaceutical companies. The software maker's installed base also includes several hospitals.