Artificial Intelligence and Computational Statistics Platform for Biosimilar Subvisible Characterization

Description

The Food and Drug Administration�s Office of Product Quality Research (OPQR) require a machine learning (ML/AI) and computational statistics platform with associated services to detect and classify protein aggregates in biosimilar drug products. This capability will support a feasibility study assessing the utility of artificial intelligence/machine learning and computational statistical analysis for biosimilar comparability assessment, quality assessment, and quality surveillance. The platform: � Shall combine machine learning to generate morphological fingerprints of protein aggregates � Shall generate morphological fingerprints specific to product and underlying stress or mechanism of aggregation � Shall be able to differentiate particles from different stress types, the product, and container closure system. � Shall combine computational statistics and neural network-based metric learning to characterize heterogeneous suspensions of subvisible particles (those <100 microns) in biologic and biosimilar drug products � Shall be compatible with Flow Imaging and Backgrounded Membrane Imaging data with no prior requirement for image processing � Shall combine computational statistics and neural network-based metric learning to characterize and predict potential root cause of particle formation in biosimilar drug products � Shall provide quantitative data on the aggregate and particle population inherent in biopharmaceuticals as opposed to simple size and count method used to characterize particles in drug solutions. � Shall employ statistical analysis tools such as Euclidian distance, similarity score based on the Kolmogorov-Smirnov test or superior statistical tool � Shall be a trusted, acceptable model used by the biopharmaceutical industry � Shall have demonstrable experience and prior publications in applying supervised and unsupervised machine learning approaches to classify visible and subvisible particle images in biologics � Shall compensate for optical phenomenon at different length scales � Shall allow visual examination of at least the twenty nearest images to any point selected on the Fingerprint. � Training provided to DPQR staff on application of AI/ML for particle classification and interpretation of results from AI particle classification approaches for product quality analysis The Government will award a contract resulting from this solicitation to the responsible quoter as a fixed?price contract on the lowest price technically acceptable (LPTA) evaluation method. Award will be made on the basis of the lowest evaluated price meeting or exceeding the non?cost factor (technical conformance to the requirements of the solicitation). The Quoter�s initial quotation shall contain the Quoter�s best terms from a price standpoint. Failure to demonstrate meeting any of the requirements will result in a rating of technically unacceptable and will not be considered for award. The following factors shall be used to evaluate quotes: � Total price. � Technical features meeting/exceeding requirements specified. For further details, please review the attached RFQ_FDA-75F40126Q00142 document.