State-of-the-art machine learning methods for predicting associations between biological objects rely on representing those objects as points in a low-dimensional metric space. Recent decade has seen the development of improved methods that utilize hyperbolic space as the embedding space of biological objects. In particular, the matrix factorization method carried out in the hyperbolic space results in more accurate embeddings and more accurate biological relationship inference when compared to methods that utilize the Euclidean latent space geometry. Unfortunately, optimization process in hyperbolic space is notoriously hard due to numerical instabilities arising from rough and rugged surfaces of error functions. In this short paper, we build upon the algorithm for matrix factorization in the Lorentzian space and present a heuristic approach for controlling the erratic behavior of hyperbolic gradient. As the main result of this study, we demonstrate that, in the context of drug-target interactions, the accuracy of hyperbolic matrix factorization compares favorably to the accuracy of methods based on artificial neural networks.