Control of mammalian cell-based devices with genetic programming

Signal integration and integral feedback control with biochemical reaction networks

Biochemical reaction networks perform a variety of signal processing functions, one of which is computing the integrals of signal values. This is often used in integral feedback control, where it enables a system's output to respond to changing inputs, but to then return exactly back to some pre-determined setpoint value afterward. To gain a deeper understanding of how biochemical networks are able to both integrate signals and perform integral feedback control, we investigated these abilities for several simple reaction networks. We found imperfect overlap between these categories, with some networks able to perform both tasks, some able to perform integration but not integral feedback control, and some the other way around. Nevertheless, networks that could either integrate or perform integral feedback control shared key elements. In particular, they included a chemical species that was neutrally stable in the open loop system (no feedback), meaning that this species does not have a unique stable steady-state concentration. Neutral stability could arise from zeroth order decay reactions, binding to a partner that was produced at a constant rate (which occurs in antithetic control), or through a long chain of covalent cycles. Mathematically, it arose from rate equations for the reaction network that were underdetermined when evaluated at steady-state.

Teaching systematic, reproducible model development using synthetic biology

We present an educational unit to teach computational modeling, a vital part of chemical engineering curricula, through the lens of synthetic biology. Lectures, code, and homework questions provide conceptual and practical introductions to each computational method involved in the model development process, along with perspectives on how methods can be iterated upon to arrive at a final model. Ultimately, this content can be applied broadly to address questions in synthetic biology and classical chemical engineering.

Immune System Programming: A Machine Learning Approach Based on Artificial Immune Systems Enhanced by Local Search

The foundation of machine learning is to enable computers to automatically solve certain problems. One of the main tools for achieving this goal is genetic programming (GP), which was developed from the genetic algorithm to expand its scope in machine learning. Although many studies have been conducted on GP, there are many questions about the disruption effect of the main GP breeding operators, i.e., crossover and mutation. Moreover, this method often suffers from high computational costs when implemented in some complex applications. This paper presents the meta-heuristics programming framework to create new practical machine learning tools alternative to the GP method. Furthermore, the immune system programming with local search (ISPLS) algorithm is composed from the proposed framework to enhance the classical artificial immune system algorithm with the tree data structure to deal with machine learning applications. The ISPLS method uses a set of breeding procedures over a tree space with gradual changes in order to surmount the defects of GP, especially the high disruptions of its basic operations. The efficiency of the proposed ISPLS method was proven through several numerical experiments, including promising results for symbolic regression, 6-bit multiplexer and 3-bit even-parity problems.