Fast activation maximization for molecular sequence design

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.

Computational Sequence Design Techniques for DNA Microarray Technologies

In systems biology and biomedical research, microarray technology is a method of choice that enables the complete quantitative and qualitative ascertainment of gene expression patterns for whole genomes. The selection of high quality oligonucleotide sequences that behave consistently across multiple experiments is a key step in the design, fabrication and experimental performance of DNA microarrays. The aim of this chapter is to outline recent algorithmic developments in microarray probe design, evaluate existing probe sequences used in commercial arrays, and suggest methodologies that have the potential to improve on existing design techniques.

A Molecular Solution to the Three-Partition Problem

Given a set of numbers, the three-partition problem is to divide them into disjoint triplets that all have the same sum. The problem is NP-complete. This paper presents an algorithm to solve this problem using the biomolecular computing approach. The algorithm uses a distinctive encoding technique that depends on the numbers values which omits the need to an adder to find the sum. The algorithm is explained and an analysis of its complexity in terms of time, the number of strands, number of tubes, and the longest library strand used is presented. A simulation of the algorithm is implemented and tested. This algorithm further proves the ability of molecular computing in solving hard problems.

Computational Sequence Design Techniques for DNA Microarray Technologies

In systems biology and biomedical research, microarray technology is a method of choice that enables the complete quantitative and qualitative ascertainment of gene expression patterns for whole genomes. The selection of high quality oligonucleotide sequences that behave consistently across multiple experiments is a key step in the design, fabrication and experimental performance of DNA microarrays. The aim of this chapter is to outline recent algorithmic developments in microarray probe design, evaluate existing probe sequences used in commercial arrays, and suggest methodologies that have the potential to improve on existing design techniques.

Experimentally Constructing Semantic Models Based on DNA Computing

We propose a new DNA-based semantic model, constructed of DNA molecules, called asemantic model based on molecular computing(SMC). It is structured as a graph formed by the set of all (attribute, attribute value) pairs contained in the set of represented objects, plus a tag node for each object. Each path in the network, from an initial object-representing tag node to the terminal node, represents the object named on the tag. Inputting a set of input strands the forms object-representing dsDNAs via parallel self-assembly from encoded ssDNAs representing both attributes and attribute values (nodes), as directed by ssDNA splitting strands representing relations (edges) in the network. The success of experiments in constructing a small test model demonstrates that our proposed model suitably represents knowledge to storing vast amounts of information at high density.

On combinatorial DNA word design

We consider the problem of designing DNA codes, namely sets of equi-length words over the alphabet [A, C, G, T] that satisfy certain combinatorial constraints. This problem is motivated by the task of reliably storing and retrieving information in synthetic DNA strands for use in DNA computing or as molecular bar codes in chemical libraries. The primary constraints that we consider, defined with respect to a parameter d, are as follows: for every pair of words w, x in a code, there are at least d mismatches between w and x if w not equal x and also between the reverse of w and the Watson-Crick complement of x. Extending classical results from coding theory, we present several upper and lower bounds on the maximum size of such DNA codes and give methods for constructing such codes. An additional constraint that is relevant to the design of DNA codes is that the free energies and enthalpies of the code words, and thus the melting temperatures, be similar. We describe dynamic programming algorithms that can (a) calculate the total number of words of length n whose free energy value, as approximated by a formula of Breslauer et al. (1986) falls in a given range, and (b) output a random such word. These algorithms are intended for use in heuristic algorithms for constructing DNA codes.