Stem cell differentiation platform utilizing Bayesian machine learning

在2032年，我们采取了三个步骤来实现我们的目标，即实现用于诱导心脏分化的小分子的硅设计我们开发了一种基于范德华体积形状和亲水性信息的新型分子描述符。在神经网络模型中使用这些描述符，我们对小分子的心脏分化效果取得了较好的预测准确率（真阳性率为77%）。我们开发了一个基于分子片段库的硅内生成小有机化合物的新代码。该代码生成的候选化合物可以通过上面描述的神经网络进行测试。[3]我们在实验室成功地实现了心脏分化方案，可用于测试上述预测的化合物。

During 2032, we made three steps towards achieving our goal of realising in-silico design of small molecules for inducing cardiac differentiation.[1] We developed a new type of molecule descriptor based on van der Waals volume shapes and hydrophilicity information. By using these descriptors in neural network model, we achieved good predictive accuracy for the cardiac differentiation effect of small molecules (true positive rate = 77 %). [2] We developed a new code for in-silicon generation of small organic compounds based on a molecule fragment library. The candidate compounds generated by this code can be tested by the neural network described above. [3] We succeeded to implement a cardiac differentiation protocol in our laboratory, which can be used to test the compounds predicted above.