Fast activation maximization for molecular sequence design

Abstract

Background Optimization of DNA and protein sequences based on Machine Learning models is becoming a powerful tool for molecular design. Activation maximization offers a simple design strategy for differentiable models: one-hot coded sequences are first approximated by a continuous representation, which is then iteratively optimized with respect to the predictor oracle by gradient ascent. While elegant, the current version of the method suffers from vanishing gradients and may cause predictor pathologies leading to poor convergence. Results Here, we introduce Fast SeqProp, an improved activation maximization method that combines straight-through approximation with normalization across the parameters of the input sequence distribution. Fast SeqProp overcomes bottlenecks in earlier methods arising from input parameters becoming skewed during optimization. Compared to prior methods, Fast SeqProp results in up to 100-fold faster convergence while also finding improved fitness optima for many applications. We demonstrate Fast SeqProp’s capabilities by designing DNA and protein sequences for six deep learning predictors, including a protein structure predictor. Conclusions Fast SeqProp offers a reliable and efficient method for general-purpose sequence optimization through a differentiable fitness predictor. As demonstrated on a variety of deep learning models, the method is widely applicable, and can incorporate various regularization techniques to maintain confidence in the sequence designs. As a design tool, Fast SeqProp may aid in the development of novel molecules, drug therapies and vaccines.

Efficient algorithms for designing maximally sized orthogonal DNA sequence libraries

“…Since then, graph representations of sequences have become widespread as descriptive tools in bioinformatics, used to reconstruct naturally occurring biological sequences. In modern molecular biology and bioengineering, where the design of synthetic biological systems is fundamentally intertwined with the characterization of natural biological systems, there is growing interest in sequence representations amenable to design tasks (16, 17). However, outside of highly specialized applications (18, 19), graph representations of sequences are far less commonly used in design contexts.…”

Now What Sequence? Pre-trained Ensembles for Bayesian Optimization of Protein Sequences

“…The standard solution to this is the Gumbel-Softmax Trick. 15 We used a slightly different approach introduced by Linder and Seelig [13] , which is to simply add a trainable layer normalization that can trainably affect the mean and variance of logits. 66 Recently, Daulton et al [67] showed that BO with probabilistic reparameterization ,like Equation 8, will converge to the true maximum of the acquisition function.…”

“…The standard solution to this is the Gumbel-Softmax Trick. 15 We used a slightly different approach introduced by Linder and Seelig [13] , which is to simply add a trainable layer normalization that can trainably affect the mean and variance of logits. 66 Recently, Daulton et al .…”

“…First, BO requires a gradient of the surrogate model with respect to its input, but sequence models have discrete integers as input. We use probabilistic reparameterization proposed in Linder and Seelig [13] , which is similar to the Gumbel softmax-trick 14,15 to enable gradient ascent of a pretrained sequence model. The second reason is that pretrained models do not have uncertainties to compute probabilities in the BO algorithm.…”

Machine Learning for Designing Next-Generation mRNA Therapeutics

“…BMC Bioinformatics 2021, 22, 510. 2 Fast SeqProp is a computational design algorithm based on activation maximization that can be combined…”