Over the past decade, low graduation and retention rates have plagued higher education institutions. To help students graduate on time and achieve optimal learning outcomes, many institutions provide advising services supported by educational technologies. Accurate grade prediction is an integral part of these services such as degree planning software, personalized advising systems and early warning systems that can identify students at-risk of dropping from their field of study. In this work, we present next-term grade prediction models based on students' cumulative knowledge and co-taken courses. The proposed models are based on a matrix factorization framework and incorporate a co-taken course interaction function to learn the influence from the co-taken courses on the target course. The co-taken course interaction function is formed by a neural network, which takes the knowledge difference between the co-taken courses and the target course as input, and outputs an influence value that will be used to predict students' grades on the target course. The experimental results on various datasets from a U.S. University demonstrate that the proposed models significantly outperform competitive baselines across different test sets. Furthermore, we analyze the proposed models' performance with different numbers of co-taken courses as well as different numbers of co-taken course subjects, and highlight with an application case study how a student might make decisions related to selection of courses. The codes are available at https://github.com/Zhiyun0411/EDM. [For the full proceedings, see ED599096.]