Applying molecular similarity used for evaluating the accuracy of retention index predictions in gas chromatography using deep learning