Abstract The development of new protein biosensors has for the most part been dependent on finding a protein that is already responsive to a known effector. While rational design and directed evolution methods exist for altering the effector specificity of transcription factors, these methods are in general complex and slow, and have failed to solve the more general problem of identifying new protein biosensors at will. In particular, it is often difficult to find a receptor that is both sensitive and specific for a given end product or intermediate, and even when efforts to generate new sensors are successful, they generally recognize effectors that are structurally quite similar to their natural counterparts. In particular, for virtually all industrially and medically useful terpenes there exists no corresponding biosensor. We now propose to develop a combined computational and directed evolution method that should allow us to proceed from any of a wide variety of ‘generalist’ repressors to create highly sensitive and specific biosensors for a structurally diverse range of terpenes and terpenoids for which no biosensors are currently known. To this end, we have developed a novel directed evolution method for altering biosensor specificities, and propose to synergize these with powerful machine learning tools for improving protein function. Extensive Preliminary Results show that the TetR family of transcription factors can be readily manipulated to take on new effector specificities, and that machine learning can be used to improve the function of a wide variety of proteins. We now further propose to identify semi-specific transcription factors as starting points for biosensor design and evolution (Aim 1); use neural network approaches to predict new sensor specificities (Aim 2); and refine these predictions via directed evolution and high-throughput screening (Aim 3).