Predicting Seizure Liability of Pharmaceuticals Using Machine Learning on Ca Imaging

2023 Fiscal Year Final Research Report

• Project/Area Number: 22K17998
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 62010:Life, health and medical informatics-related
• Research Institution: Tohoku Institute of Technology
• Principal Investigator: Matsuda Naoki 東北工業大学, 工学部, 助教 (80909490)
• Project Period (FY): 2022-04-01 – 2024-03-31
• Keywords: 機械学習 / 脳・神経 / in vitro / カルシウムイメージング / 神経毒性

Outline of Final Research Achievements

This study aimed to develop a method for predicting seizure risk using machine learning on high-throughput Ca imaging data. To accurately detect Ca oscillations, we explored a Ca oscillation detection method using SVM, and developed a model capable of detecting oscillations with 97% accuracy. Next, we created images of the detected oscillation waveforms and developed a seizure toxicity prediction model using a CNN. The CNN model predicted 5 types of negative compounds and 27 types of seizure-positive compounds with 87% accuracy and detected the toxicity of seizure-positive compounds in a dose-dependent manner. There were no false detections of negative compounds, demonstrating the effectiveness of this method for predicting seizure toxicity using Ca imaging.

Free Research Field

生体医工学

Academic Significance and Societal Importance of the Research Achievements

ハイスループット性の高いCaイメージング法による痙攣毒性予測法は、創薬開発の探索段階における毒性検出を可能とし、化合物の選別、薬剤の有効性、副作用の評価や作用機序、リード化合物におけるリスクの順位付け、リスクを回避する為の化合物の修飾等が可能となり、創薬開発におけるコストと時間の大幅な削減につながると考えられる。また、本開発技術は、痙攣毒性のみならず、他の神経毒性および薬効への適用も期待できる。