Unifying Likelihood-free Inference with Black-box Optimization and Beyond

Abstract

Black-box optimization formulations for biological sequence design have drawn recent attention due to their promising potential impact on the pharmaceutical industry. In this work, we propose to unify two seemingly distinct worlds: likelihood-free inference and black-box sequence design, under one probabilistic framework. In tandem, we provide a recipe for constructing various sequence design methods based on this framework. We show how previous drug discovery approaches can be "reinvented" in our framework, and further propose new probabilistic sequence design algorithms. Extensive experiments illustrate the benefits of the proposed methodology.

Proximal Exploration for Model-guided Protein Sequence Design

“…Brookes & Listgarten (2018) and Brookes et al (2019) use adaptive sampling to generate high-quality mutants for batch Bayesian optimization. A probabilistic framework is proposed by Zhang et al (2021) that unifies black-box sequence design and likelihood-free inference and presents a methodology to develop new algorithms for sequence design.…”

Accelerating Bayesian Optimization for Biological Sequence Design with Denoising Autoencoders

“…Nigam et al (2019) combine genetic algorithms and neural networks to optimize multi-objective chemical properties. The LaMBO architecture is inspired by "iterative resampling" approaches to sequence design and machine translation (Gligorijevic et al, 2021;Lee et al, 2018;Zhang et al, 2021).…”

Bayesian max-product expectation maximization algorithm for structured sparse signals reconstruction

“…Protein sequence design has been studied with a wide variety of methods, including traditional directed evolution (Arnold, 1998;Dalby, 2011;Packer & Liu, 2015;Arnold, 2018) and machine learning methods. The mainly used machine learning algorithms include reinforcement learning (Angermueller et al, 2019;Jain et al, 2022), Bayesian optimization (Belanger et al, 2019;Moss et al, 2020;Terayama et al, 2021), search using deep generative models (Brookes & Listgarten, 2018;Brookes et al, 2019;Madani et al, 2020;Kumar & Levine, 2020;Das et al, 2021;Hoffman et al, 2022;Melnyk et al, 2021;Ren et al, 2022) adaptive evolution methods (Hansen, 2006;Swersky et al, 2020;Sinai et al, 2020) as well as likelihood-free inference (Zhang et al, 2021).…”