C-COUNT: a convolutional neural network-based tool for automated scoring of erythroid colonies

Area of Science:

• Hematology
• Computational Biology
• Biotechnology

Background:

• Colony-formation assays (CFAs) are crucial for evaluating erythroid and hematopoietic progenitors in research and diagnostics.
• Manual counting of colonies in CFAs is time-consuming, inconsistent, and prone to bias.
• Existing methods like flow cytometry and single-cell transcriptomics do not fully replace functional progenitor assays.

Purpose of the Study:

• To develop and validate an automated tool, C-COUNT, for accurate scoring of the colony-forming-unit-erythroid (CFU-e) assay.
• To improve the efficiency, reliability, and objectivity of CFU-e assay analysis.
• To enable high-throughput screening applications in hematology research.

Main Methods:

• Development of a convolutional neural network (CNN)-based tool named C-COUNT.
• Training and validation of the CNN using images from automated microscopy of CFU-e assays.
• Comparative analysis of C-COUNT performance against experienced human scorers.
• Evaluation of C-COUNT in response to erythropoietin and genotoxic agents.

Main Results:

• C-COUNT reliably identifies and quantifies CFU-e colonies, even in mixed cultures with myeloid colonies and cell aggregates.
• The performance of C-COUNT is equivalent or superior to that of experienced scientists in colony identification.
• The tool accurately assesses CFU-e progenitor responses to varying erythropoietin concentrations and genotoxic agents.

Conclusions:

• C-COUNT transforms the traditional CFU-e CFA into a rigorous, efficient, and objective assay.
• The automated tool enhances the utility of CFAs for hematology research and clinical diagnosis.
• C-COUNT has significant potential for high-throughput screening of erythropoietic factors and therapeutic agents.