Johns Hopkins APL Licenses Powerful Machine-Learning Tools to BullFrog AI

The Johns Hopkins Applied Physics Laboratory (APL), in Laurel, Maryland, has licensed two powerful machine-learning tools — named Prometheus and Seagull — to BullFrog AI Inc. The agreement enables the Maryland-based biotechnology research firm to use the tools to analyze clinical trial data and potentially make life-saving therapies and treatments available to patients more quickly.

One of the major challenges in developing new therapeutics is efficiently integrating complex and high-dimensional data generated at each stage of development to reduce any risks in subsequent stages of the development process. Artificial intelligence and machine learning have emerged as digital solutions to help solve this problem.

Prometheus includes a comprehensive library of APL-invented probabilistic models, one of which — Random Subspace Mixture Models — recently beat 10 competing algorithms in detecting anomalies in an extensive benchmarking study done using 12 open-source data sets. Prometheus also includes graph algorithms that have been successfully used to analyze large-scale network resilience problems on multimodal global networks, such as global logistics and communication networks.

Prometheus can also be used to construct networks from multimodal and potentially incomplete data to discover relationships between entities and attributes (such as patient medical records) through link inference. One of the link centrality metrics in Prometheus (link cohesion) was recently used to prune a noisy network to discover hidden clusters more accurately than existing alternatives are able to.

Seagull provides a comprehensive library of multivariate time series analyses. Multivariate time series pertains to a sequence of observations collected about the attributes and behavior of “actors,” such as a person, computer or company. Seagull can enrich time series data by fusing it with open-source data as well as calculated behavioral features.

Besides single-actor analysis, Seagull also provides efficient implementations of time series correlation and clustering capabilities that allow it to rapidly discover associations between entities in near linear time — a computationally challenging task.

APL has used Seagull and Prometheus in various sponsored projects for the detection of anomalous activities, noted Cetin Savkli, a chief scientist in APL’s Asymmetric Operations Sector and the inventor of Prometheus and Seagull.

Both tools evolved from APL’s Socrates software, a graph analytics application that discovers patterns and relationships in large, complex sets of data. Socrates grew out of a need across many Laboratory projects to analyze large-scale, “messy” data, said Savkli, who also invented Socrates.

BullFrog AI was granted a license to Socrates in 2017. “The collaboration with APL continues to enrich BullFrog AI’s platform with the aim of using AI to advance medicine and enrich patient lives,” said Vin Singh, founder and chief executive officer of BullFrog AI.

“Over the last five years, probabilistic models, graph algorithms, and time series analysis capabilities have been significantly expanded and deepened with new inventions and are now housed under Prometheus and Seagull libraries, which succeed Socrates,” Savkli said.

APL has a long-standing leadership role in large-scale analytics based on its creation of novel software tools that can be integrated into systems to tackle the challenge of machine learning at scale.

“APL is excited to enable BullFrog AI to accelerate the creation of life-saving treatments by giving them access to APL’s powerful machine-learning tools,” said Norma Lee Todd, head of APL’s Tech Transfer. “This is an opportunity to salvage failed drug candidates or repurpose existing drugs for new indications to save lives.”

APL Tech Transfer offers companies the ability to access cutting-edge innovations made at the Lab to benefit society and improve the lives of people throughout the world, Todd added.