Abstract
A protein is defined by its amino acid sequence. This sequence and environmental factors shape a protein's 3D structural landscape, which is crucial for the protein's function and activity. Protein design aims to develop novel protein sequences or modify existing ones to perform specific functions or have desired protein properties. The protein sequence space is exponentially large, making protein sequence design a tough problem. This problem can be simplified by considering a backbone conditional protein sequence design that factorizes the design problem into two parts: protein backbone design and backbone-dependent sequence design. This allows for a more efficient search over the sequence space for desired structural features. In this review, we discuss when backbone conditional sequence design is possible and how to assess the performance of different design methods, training data, symmetric design, and the combination of unconditional and conditional sequence models.