Functional classification based on protein sequences has become a critical challenge due to the exponential growth of protein data in biological databases. The remarkable diversity within homologous sequences often obscures an array of distinct functional activities, many of which cannot be reliably inferred through conventional methods. Accurate identification and characterization of these functions are essential, both for unraveling fundamental evolutionary processes in living organisms and for harnessing biotechnological potential. I will discuss how deep learning approaches, by re-envisioning sequence space, facilitate the accurate functional classification of proteins directly from their sequences. Such computational strategies not only enhance our understanding of protein evolution but also open avenues for designing innovative functional switches and generating novel, high-efficiency artificial proteins.

Membrane systems represent a computational model that operates in a distributed and parallel manner, inspired by the behavior of biological cells.
These systems feature objects that transform within a nested membrane structure.
This research concentrates on a specific type of these systems, based on cellular symport/antiport communication of chemicals.

Results in the literature show that systems of this type that also allows cell division can solve PSPACE problems.
In our study, we investigate systems that use membrane separation instead of cell division, for which only limited results are available.
Notably, it has been shown that any problem solvable by such systems in polynomial time falls within the complexity class P^#P.

By implementing a system solving MIDSAT, a P^#P-complete problem, we demonstrate that the reverse inclusion is true as well, thus providing an exact characterization of the problem class solvable by P systems with symport/antiport and membrane separation.

Moreover, our implementation uses rules of length at most three.
With this limit systems were known able to solve NP problems, whereas limiting the rules by length two is characterizes P.

This paper studies the effects of interactivity on molecular computation, specifically in Step Chemical Reaction Networks model (Step CRNs), by adding the ability for a user to interact with the system by selecting which species to add at each step, or by having some control over which reactions execute. The two proposed variants are Interactive CRNs and Randomized Interactive CRNs. We show that in Interactive CRNs, even when restricted to void (deletion-only) rules of relatively small size, if a user can decide which species to add at each step based on the configuration, reachability is PSPACE-complete when bounded and EXPTIME-hard when unbounded. In Randomized Interactive CRNs, we prove that reachability with void rules is PSPACE-complete.

Central Pattern Generators (CPGs) are neural circuits capable of autonomously producing rhythmic output patterns without requiring rhythmic input. They maintain stable phase relationships among constituent oscillators, adapt to sensory feedback and descending modulation, and exhibit resilience to perturbations. This work presents a novel hybrid Analog-Digital CPG architecture that leverages the continuous, low-latency dynamics of analog oscillators alongside the flexibility, reconfigurability, and precision of digital control. The proposed system features tunable parameters and a modular design, enabling realtime adaptation to sensory inputs and environmental conditions. This approach provides a versatile framework for generating robust, adaptable gait patterns, with applications spanning autonomous robotics, soft robotics, and human-assistive technologies.

This paper describes an efficient variation of a well-known technique for converting a general-purpose analog computer (GPAC) to a chemical reaction network (CRN). If an input system has n variables, the existing dual-railing technique requires 2n variables in the resulting system. As the resulting CRNs are often fed into downstream processes which may themselves be exponential in the number of variables, it is frequently worthwhile to try to minimize the number of variables generated in this conversion. This paper presents a technique that results in fewer than n new variables, by rewriting only those whose positive terms are ‘infected’ by ill-formed variables.

Tree structures appear in many fields of the life sciences. For example, phylogenetic trees are an important tool for understanding evolution, and cell lineage trees are at the center of understanding cell division and differentiation. Trees are also used to represent RNA secondary structures, which directly relate to the function of non-coding RNAs. Recent developments in sequencing technology and artificial intelligence have yielded numerous biological data that can be represented with tree structures. Methods for tree structure data analytics are in need. Tree polynomials provide a computationally efficient, interpretable and comprehensive way to encode tree structures as matrices, which are compatible with most data analytics tools. Machine learning methods based on the Canberra distance between tree polynomials have been introduced to analyze phylogenies and nucleic acid structures. In this paper, we compare the performance of different distances in tree clustering methods based on a tree distinguishing polynomial. We also implement two basic autoencoder models for clustering trees using the polynomial. We find that the distance based methods with entry-level normalized distances have the highest clustering accuracy among the compared methods.

The radical pair mechanism, which is considered to aid avian navigation in the geomagnetic field, is analyzed from the information-theoretic perspective. Previously set within the information-theoretic framework and shown to fulfill a necessary condition for emergent functionality, it is revisited hereby and further investigated according to the recent developments of information theory for describing emergent phenomena. Some initial simulations indicate that the entropic relations which reveal the structure of information within the investigated system hold within the addressed framework and provide directions for further investigation.