I am interested in understanding biological phenomena through general principles and theoretical models. Instead of studying biological systems for which traditional physics models and methods are applicable, I would like to see how biology offers new types of phenomena and questions that are different from usual physical systems, which require new approaches and may expand the scope of physics. Compared to other forms of physical systems (such as condensed matter), biological organisms (or “living matter”) are special in many aspects. The unique aspects of living organisms offer great opportunities for thinking about the nature of biology and its connection to physics.

Phenotypic heterogeneity within populations

Different organisms exhibit phenotypes, such as morphological and behavioral traits, that are characteristic of their species. However, even within a population of the same species, the phenotypes may vary substantially among individuals. This is in contrast to systems often studied in physics, which are made of identical and indistinguishable particles. Phenotypic heterogeneity within a population can play an important role in the adaptation and evolution of the species.

• bet-hedging: the adaptive value of information and memory

Phenotypic variation can help a population adapt to fluctuating environments. By diversifying into subpopulations of different phenotypes that are favorable under different environmental conditions, the population can survive unpredictable environmental changes. The more uncertain the environment is, the more diverse the population needs to be. This “bet-hedging” strategy is found in many organisms, such as persister cells in bacteria, seed dormancy in plants, and defense mechanisms in animals. Some organisms utilize environmental cues to preferably express phenotypes that are favorable in more likely environmental conditions. The benefit of such cues for population growth is mathematically connected to the amount of information about the environment contained in the cues. Furthermore, we showed that certain internal states of the organisms can serve as internal cues because they are correlated with the environment as a result of selection. The more memory these states contain about the past environment, the better the organisms can adapt to temporally structured environments [ref].

• mimicry: the statistical limit of resemblance and distinguishability

Coloration patterns on animal bodies are known to vary significantly between individuals, and can even serve as “fingerprints” for identifying individuals. Such patterns are important for many biological processes, including camouflage, warning signals, mating displays, etc. For example, in Batesian mimicry, a nontoxic species can display patterns similar to a toxic species to fool the predator and avoid attack. However, the level of resemblance varies from one species to another, and from one individual to another even within the same species. The extent to which the patterns vary within a species has not been well characterized. What consequences can such pattern variations have on the evolution of the species? We are developing machine-learning algorithms that can infer low-dimensional representations of complex coloration patterns. Such representations will be used as a morphological space to quantify visual traits and allow the study of variation among individuals and between species. We will be able to study evolutionary questions, such as those involving mimicry: Do mimics look more similar in the eye of the predators? Does intraspecific variation affect the distinguishability of the mimicry species?

Interaction modification by third parties

Biological systems are often described by pairwise interactions between components, such as protein interactions within a cell or species interactions in an ecosystem. However, unlike physical systems where the interactions are determined by fundamental forces with universal coupling constants, the interaction strengths between biological components can be modified by third parties, such as the allosteric regulation of protein interactions. That is because biological components have complex internal structures, and the effective strengths of their interactions can change if the internal structures are altered.

• trait-mediated effects in ecology

Traditional models of ecosystems assume a constant interaction strength between a pair of species. However, individuals of the same species vary in traits that can affect how they interact with the environment or other species. The distribution of traits may shift under the influence of other species, leading to the modification of interaction strengths. As an example, we analyzed an ecosystem with two predators and a shared prey [ref], which would result in the competitive exclusion of one predator if the interaction strengths are fixed. We showed that when there is trait variation within the prey that can shift dynamically, the ecosystem can stabilize at a new equilibrium where all three species coexist. We also uncovered trends that are unexpected in models with constant interaction strengths, such as the facilitation of one predator by the other (i.e., the abundance of one predator is increased when the other is present) and the promotion of the prey by one predator (i.e., the abundance of the prey is increased when the predator is present). These “abnormal” trends have broad consequences on many types of ecosystems and larger networks.

• robust retrieval of dynamic sequences

Many biological systems are able to dynamically rearrange their components through a sequence of configurations in order to perform their functions, such as neurons altering their synaptic strengths to generate a sequence of activity patterns, or protein assemblies rearranging their composition to go through several stages of processing substrates. Such dynamic processes have been studied using network models that sequentially retrieve a set of stored patterns by controlling inputs to the network components (“input modulation”). In contrast, we introduced a new class of models that can retrieve dynamic sequences by modifying the interactions among the components (“interaction modulation”) [ref]. We showed that interaction modulation models can robustly retrieve patterns with different activity levels and have a much larger dynamic storage capacity. Such interaction modulation may be a new paradigm for modeling complex collective dynamics.

Internal representation of the environment

Organisms are “evolved matter” whose form and function have been shaped by the environment that they have encountered. Compared to inorganic matter that reacts to external conditions in simple aimless ways, organisms exhibit far more complex responses to environmental stimuli, as seen in remarkable examples of morphological adaptations and behavioral strategies. These sophisticated responses reflect a knowledge of the environment that has been accumulated over the course of evolution as well as during individuals’ lifetime. Such knowledge can be described as an internal representation of the environment that organisms rely on to process information and react accordingly.

• epigenetic inheritance and evolutionary learning

In the example of bet-hedging, a population benefits from phenotypic diversity if the distribution of phenotypes matches the variability of the environment. In this sense, the amount of diversity maintained in the population is a representation of the environmental statistics. To attain the optimal phenotype diversity, however, organisms must overcome an apparent mismatch in timescale: the individual lifespan is too short for gathering information about the environmental statistics. We proposed an “evolutionary learning” process, by which the phenotype distribution of the population can be dynamically adapted [ref]. It happens if each organism alters the phenotype distribution of its offspring to increase the frequency of its own phenotype, analogous to Hebbian learning in neuroscience. This process may be realized through various mechanisms of epigenetic inheritance, which are actively studied in biology.

• phenotypic plasticity and environment-to-phenotype mapping

Phenotypic plasticity is the ability of organisms to develop alternative phenotypes depending on environmental cues. It is an important aspect of evolution, because in such cases the environment plays not only a selective but also a formative role. We proposed to describe phenotypic plasticity by a mapping from environmental cues to phenotypic traits, which is an abstraction of many possible mechanisms for gene regulation. Using this “environment-to-phenotype mapping”, we showed that the evolutionary benefit of phenotypic plasticity is determined by the predictability of future environment at the time of development [ref]. Moreover, depending on the noise in environmental cues and the strength of natural selection, phenotypic plasticity can give rise to different strategies for adapting to varying environments, including “unvarying” (where organisms express a constant “generalist” phenotype), “tracking” (where organisms match phenotypes to the environment), and “bet-hedging” [ref].