ARTIFICIAL INTELLIGENCE PLATFORM – Advancing Precision Medicine With bfLEAP™: Next Generation AI for Drug Development


INTRODUCTION

In the ever-evolving landscape of drug development, the pur­suit of precision medicine has gained tremendous momentum. This approach aims to revolutionize healthcare by tailoring treat­ments to individuals based on their unique genetic makeup, lifestyle, and environmental factors.

BullFrog AI (NASDAQ: BFRG) is a digital technology com­pany that is leveraging the potential of artificial intelligence (AI) and machine learning (ML) to advance this field and improve suc­cess rates in drug development. AI and ML have emerged as game-changers in the drug discovery and development industry, enabling computers to analyze vast amounts of complex data, identify patterns, and extract valuable insights for drug developers and clinicians. These technologies hold promise for revolutioniz­ing patient care and advancing more effective therapies to market faster. Why is this so important?

  • Getting a drug to market can take close to a decade, with a cost of $0.8 to $2.8 billion, while patients are desperately wait­ing for treatments that often don’t come.1
  • Almost 90% of drugs fail at some point along the way.2

BullFrog AI is focused on changing this narrative using bfLEAP™, an AI platform exclusively licensed from the prestigious Johns Hopkins Applied Physics Laboratory (APL) for biological and chemical pharmaceutical therapeutics applications. bfLEAP is a robust platform that can help to rapidly detect anomalies and uncover hidden associations within patient data, making it a pow­erful tool with the potential to help researchers predict drug re­sponses and identify patient subgroups.

With an advanced technical architecture, bfLEAP solves nu­merous scalability and flexibility challenges, facilitating the com­prehensive analysis of diverse and complex data sets. Using unsupervised ML algorithms and proprietary clustering tech­niques, the platform identifies potentially meaningful and under­standable information, paving the way for the creation of personalized treatments, optimized clinical trials, disease progres­sion predictions, and drug target identification.

The following explores the technical architecture and capa­bilities of bfLEAP, delving into its ML algorithms, proprietary clus­tering techniques, and visually understandable outputs. Additionally, some of the potential use cases for the platform in drug development and discuss future developments that hold tremendous promise are examined.

TECHNICAL ARCHITECTURE – HIGH-THROUGHPUT, DATA-AGNOSTIC PROCESSING POWER

The true strength of bfLEAP lies in its highly efficient archi­tecture. The platform is capable of handling either small, homo­geneous data sets or large, complex data sets with equal ease. By overcoming the scalability and flexibility challenges that are commonly faced by researchers and clinicians, bfLEAP can pro­vide a path to more efficient and comprehensive data analysis.

Through the display of nodes and edges that users can explore and interact with, weighted relationships between various factors are visually illustrated, facilitating an intuitive understanding of the data.

Click image to enlarge

One of the standout features of bfLEAP is its ability to handle incomplete data sets, an all-too-common occurrence in real-world clinical settings. Whether it comes from patient dropouts, missed appointments, technical glitches, human error, or other causes, researchers are frequently faced with data gaps upon study completion. By effectively incorporating and interpreting in­complete information, this platform ensures that valuable insights are preserved.

Data agnosticism is another key attribute of bfLEAP, allowing it to integrate, analyze, and interpret heterogeneous data sources. Whether inputs consist of medical records, demographics, ge­nomic data, or real-world evidence, the platform’s ability to inte­grate and process disparate data sets is a valuable benefit to users. This versatility enables a more comprehensive analysis, un­locking new possibilities for understanding even the subtlest re­lationships.

bfLEAP excels through its powerful high-throughput process­ing capabilities, fueled by the TinkerPop™ API for parallel com­puting. This enables the rapid analysis and interpretation of data, facilitating the accelerated identification of meaningful informa­tion and potentially enabling a more efficient drug development process. This will provide researchers and clinicians actionable insights on an optimized timescale, allowing them to make data-driven decisions quickly.

UNSUPERVISED ML & PROPRIETARY CLUSTERING ALGORITHMS

The bfLEAP platform takes full advantage of the power of un­supervised ML algorithms to discover hidden patterns and rela­tionships. Unlike supervised learning methods that rely on labeled data and predefined assumptions, its unsupervised ML approach allows bfLEAP to uncover patterns organically – without human intervention. This approach ensures important insights are not overlooked due to false or limiting assumptions.

One of the significant advantages of bfLEAP is its proprietary suite of clustering algorithms, all of which are designed to uncover unknown associations between key entities, regardless of data type or use case. This versatility and flexibility allow bfLEAP to open doors in drug development that may not have previously been known to exist.

Using unsupervised ML algorithms and proprietary clustering techniques, the platform identifies potentially meaningful and understandable information.

With a total of over 200 analytic util­ities and algorithms at its disposal, the bfLEAP platform provides a comprehensive toolkit for data analysis and clustering. These algorithms have been carefully se­lected and fine-tuned over time to deliver optimal performance with maximum effi­ciency.

Another important feature of the plat­form is its anomaly detection algorithm, a random subspace mixture model (RSMM). In a rigorous benchmarking study that an­alyzed 12 open-source data sets, this al­gorithm outperformed the top 10 currently used algorithms for anomaly detection.3

EXPLAINABLE AI

Transparency and interpretability are essential qualities in AI platforms, though they are often lacking. This is particularly critical in domains such as drug develop­ment, in which the details of complex re­lationships – like drug interactions – are essential to comprehend.

bfLEAP addresses this issue through its layered processes and explainable AI ap­proach. The platform provides critical con­text to the data outputs, allowing researchers and clinicians to understand the steps involved in generating insights. This purposeful transparency is designed to not only facilitate comprehension, but to also enable sponsors to validate and trust their results.

A key component of bfLEAP’s “ex­plainable AI” approach is its graph ana­lytics output. This visual representation of relationships and correlations within the data greatly simplifies interpretation. Through the display of nodes and edges that users can explore and interact with, weighted relationships between various factors are visually illustrated, facilitating an intuitive understanding of the data. This important software feature enhances sponsor engagement, as it presents com­plex information in an easily digestible for­mat.

Yet another advantage of bfLEAP’s ex­plainable AI approach lies in its minimal custom coding requirements. Apart from the initial data cleaning and ingestion process, the platform requires few adjust­ments, ensuring shorter lead times and consistent output interpretation. This ease of use and interpretability maximizes effi­ciency and allows the platform’s full capa­bilities to be leveraged without the need for extensive programming expertise.

POTENTIAL USE CASES IN DRUG DEVELOPMENT

The versatility of the bfLEAP platform offers numerous potentially impactful use cases within the field of drug discovery and development. The following lists some specific ways in which this platform can help provide meaningful impact in the clinical research field:

Identifying Patient Subgroups to Better Predict Drug Response: By analyzing pa­tient data, bfLEAP can assist in the identi­fication of distinct subgroups based on factors such as genetic information, bio­markers, demographic information, or a variety of other patient factors and char­acteristics. This can help researchers and clinicians predict drug response within specific patient subgroups, potentially fa­cilitating personalized treatment ap­proaches and improved patient outcomes.

Informing Better Inclusion & Exclusion Criteria for Clinical Trials: Designing ef­fective clinical trials requires the inclusion of relevant patient populations while ex­cluding confounding factors. bfLEAP can aid researchers in identifying significant patient characteristics, helping them to op­timize inclusion and exclusion criteria. This helps ensure clinical trials are highly tar­geted, efficient, and yield robust results.

Predicting Disease Progression: By thor­oughly analyzing longitudinal patient data, the bfLEAP platform can help re­searchers predict disease progression pat­terns. With this knowledge in hand, clinicians may be able to intervene at ap­propriate stages more effectively, facilitat­ing early detection, proactive treatment, and improved disease management for patients. Further, they may be able to use this knowledge to better design the subse­quent phases of their program.

Identifying Pathways for Drug Applica­tion: bfLEAP analyzes genomic data and gene expression data to uncover associa­tions between networks of genes and pro­teins. By identifying these novel connections, the platform may help re­searchers discover important new targets and disease biomarkers. This enables the development of transformative therapies for unmet areas of patient need.

Drug Target Identification: By leveraging its comprehensive data analytics capabili­ties, the bfLEAP platform can assist re­searchers in identifying potential drug targets. By exploring intricate relationships and patterns within the data, the platform may help to expedite the drug discovery process and facilitate the development of highly targeted precision medicines.

Drug Repurposing and Rescue: The abil­ity to recognize previously undetected pat­terns also creates the potential to revisit the efficacy of failed drug candidates or repur­pose existing drugs for new indications.

FUTURE DEVELOPMENTS FOR bfLEAP

As the landscape of drug develop­ment evolves, BullFrog AI remains commit­ted to advancing bfLEAP and addressing this industry’s growing needs. The plat­form’s architecture is also designed to ac­commodate future developments, allowing for the seamless addition of bolt-ons to en­hance its already impressive capabilities. One key example is the implementation of natural language processing (NLP) and GraphQL integration, which can be lever­aged to mine the existing literature and further expand scientific knowledge.

This platform is continuously being re­fined and enhanced. Recent updates have streamlined the data input process, mak­ing this process faster and more efficient. This enhancement benefits sponsors who often operate on tight timelines.

Overall, by prioritizing collaboration and innovation with leading organizations and eminent scientists, BullFrog AI strives to ensure bfLEAP will remain at the fore­front of AI-driven drug development for years to come. Continued technological advancements will mean even greater so­phistication and more powerful capabili­ties that will enable researchers and clinicians to unlock new frontiers in the pursuit of advanced therapies.

With an unwavering commitment to innovation and collaboration, BullFrog AI is well positioned to help shape the future of drug development, revolutionize patient care, and transform the landscape of pre­cision medicine.

REFERENCES

  1. Wouters, O.J., McKee, M., & Luyten, J. (2020). Estimated research and development investment needed to bring a new medicine to market, 2009-2018. Jama, 323(9), 844-853.
  2. Sun, D., Gao, W., Hu, H., & Zhou, S. (2022). Why 90% of clinical drug development fails and how to improve it? Acta Pharmaceutica Sinica B, 12(7), 3049-3062.
  3. Savkli, C., & Schwartz, C. (2021). Random Subspace Mixture Models for Interpretable Anomaly Detection. arXiv preprint arXiv:2108.06283.

Dr. Thomas Hazel, BullFrog AI VP of Drug Development, has over 20 years of industry experience in R&D and business development. He most recently served as senior VP of R&D at Seneca Biopharma, overseeing the development of the company’s stem cell-based therapeutics platform. He has been granted 9 US patents and multiple foreign patents in stem cells and regenerative medicine. He earned his PhD in Genetics at the University of Illinois College of Medicine.

JT Koffenberger, BullFrog AI Chief Information Officer, has over 30 years of experience leveraging IT services for business. His range of expertise includes providing better security through custom application development and automating cumbersome business practices. He is Founder of Delmarva Group, LLC and Director of IT Architecture for Day and Zimmerman.