Homochirality is an important feature in biological systems and occurs even in inorganic nanoparticles. However, the mechanism of chirality formation and the key steps during growth are not fully understood. Here we identify two distinguishable pathways from achiral to chiral morphologies in gold nanoparticles by training an artificial neural network of cellular automata according to experimental results. We find that the chirality is initially determined by the nature of the asymmetric growth along the boundaries of enantiomeric high-index planes. The deep learning-based interpretation of chiral morphogenesis provides a theoretical understanding but also allows us to predict an unprecedented crossover pathway and the resulting morphology.

同手性是生物系统中的一个重要特征，甚至在无机纳米粒子中也存在。然而，手性形成的机制和生长过程中的关键步骤尚不完全清楚。在这里，我们根据实验结果训练细胞自动机的人工神经网络，确定了金纳米颗粒从非手性形态到手性形态的两种不同途径。我们发现手性最初是由沿着对映体高折射率平面边界的不对称生长的性质决定的。基于深度学习的手性形态发生解释提供了理论理解，但也使我们能够预测前所未有的交叉途径和由此产生的形态。