Goals

The objectives addressed by this program are wide-ranging. Due to the current efforts in the field of machine learning, it now becomes possible to grasp and solve highly complex molecular problems through computer applications. The full potential of modern ML-algorithms must be explored, and generally usable tools should be developed, simplifying and accelerating the practical research processes. These algorithms and their applications must also be understood in order to minimize the risk of unsuitable applications or erroneous predictions and gain chemical and pharmacological insights. The addressed topics can be summarized under the following 5 topics:

	Molecular representations for machine learning Development of (domain specific) molecular representations - Generally improved molecular representations - Learnable molecular representations - Representations for chemical reactions - Activity relevant molecular representations Better understanding of molecular representations
	Machine learning as a tool for theoretical and organic chemistry Prediction, understanding and interpretation of molecular properties - Improvement of current applications - Interpretation of analytical data - Prediction of photochemical properties - Prediction of different molecular properties (Quantitative) prediction of chemical reactions In silico reaction discovery De novo catalyst design
	Machine learning for medicinal chemistry and drug design Development of deep ML models for ultra-large virtual compound libraries Generative de novo design of compounds with desired drug-related properties New ML concepts for accurate and fast compound potency prediction
	Overcoming the Current Data Situation Generation and evaluation of data and datasets for machine learning - Conversion of existing data into usable corpora - Evaluation of existing and novel datasets - Generation of various relevant datasets - Combination and standardization of data Data visualization and low-dimensional embeddings Active learning strategies for predictive modeling on the basis of sparse data
	Development of machine learning tools for molecular applications Application of state-of-the-art machine learning algorithms - Transfer learning - Explainable AI - New ML-Models Interpreting “black box“ machine learning models Development of software suites and toolkits