Graph-based Complexity and Computational Capabilites of Proteinoid Spike Systems

Saksham Sharma, Adnan Mahmud, Giuseppe Tarabella, Panagiotis Mougoyannis, Andrew Adamatzky · 2025

Read Graph-based Complexity and Computational Capabilites of Proteinoid Spike Systems by Saksham Sharma, Adnan Mahmud, Giuseppe Tarabella, Panagiotis Mougoyannis, Andrew Adamatzky (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-10362.txt.

Introduction

Saksham Sharma and co-authors test whether gooey chemical computers really compute in a 2025 paper. Proteinoids, soft gel-like materials, spark with electrical spikes as ions move in and out, and sampled spikes can be translated into codes resembling logic gates. Whether that code amounts to real computing was unproven, so the team analyzed the spike patterns with graph measures and neural networks.

The patterns showed structure echoing known computing frameworks. As a demonstration, the authors built a predictor that reads sixteen numbers drawn from each spike train and classifies patterns with about seventy percent accuracy. Its core transform borrows from number theory sieves, mixing a spiral sampling step with a sign step. The work argues the gels hold genuine information-processing power.

Catalog record

CREATORSaksham Sharma, Adnan Mahmud, Giuseppe Tarabella, Panagiotis Mougoyannis, Andrew Adamatzky
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE33 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-10362

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Graph-based Complexity and Computational Capabilites of Proteinoid Spike Systems Saksham Sharma, Adnan Mahmud, Giuseppe Tarabella, Panagiotis Mougoyannis, Andrew Adamatzky Abstract Proteinoids, as soft matter fluidic systems, are computational substrates that have been recently proposed for their analog computing capabilities. Such systems exhibit oscillatory electrical activity because of cationic and anionic exchange inside and outside such gels. It has also been recently shown that this (analog) electrical activity, when sampled at fixed time intervals, can be used to reveal their underlying information-theoretic, computational code. This code, for instance, can be expressed in the (digital) language of Boolean gates and QR codes. Though, this might seem as a good evidence that proteinoid substrates have computing abilities when subjected to analog-to-digital transition, the leap from their underlying computational code to computing abilities is not well demonstrated. In this work, we analyse the electrical activity patterns of proteinoid substrates using computational methods including deep ReLU networks. Our findings suggest intriguing parallels between these patterns and certain computational frameworks, hinting at the possibility that such chemical systems might possess inherent information-processing and computational capabilities. To demonstrate the computational ability, we construct a prediction algorithm which acts as a binary classification model and extract 16-dimensional vector data from the proteinoid spike, in order to perform predictions with 70.41% accuracy. This model in its core has a unique transformation modality, inspired from number-theoretic sieve theory, and is combination of two functions: spiral sampling F 1 F_{1} and significant digit extraction F 2 F_{2} functions. The complexity of the transformed data is measured using eight distinct metrics, and effectively, using a single meta-metric . Preprint: APS/123-QED Proteinoids, made from poly(amino) acids, exhibit oscillatory analog electrical activity because of the cationic and ionic exchanges inside their gelatinous structure. This analog information can be interpreted in a digital format, by conversion to Boolean gates and QR codes. Despite this possibility, it is not yet clear, as to what can make such systems capable of performing universal computation similar to how a biological neuron does. Though relatively young in the literature on proteinoids, there are many “fluidic” soft matter systems that have been shown to have such neuron-like computing capabilities. In early 2000, Maass et al. developed Liquid State Machine consisting of a cortical microcolumn that connects the neurons randomly and is capable of universal real-time computation [ 1 ] . This model was subsequently adopted “literally” by considering water waves in a reservoir and training the system to solve XOR problem and perform speech recognition [ 2 ] . There exists plenty of reservoir computing architectures in the literature, built out of fluidic systems upon implementation of neural algorithms. A few such fluidic computing systems include the systems for signal analysis and data classification operations [ 3 ] , computation at the nonlinear regime [ 4 , 5 , 6 ] , and incorporating external force fields into the system such as acoustic fields [ 7 , 8 ] . Figure 1: (a) Voltage versus time signal of the oscillatory electrical activity inside proteinoids measured using voltage sensitive dyes (top) and the microscopic images of the proteinoid microspheres (bottom). (b) Voltage sensitive dyes (VSDs) convert the recordings into a spike-versus-time data (Dataset 1 is plotted), (c) Functions F_{1} and F_{2} are used to transform Dataset 1 into a multi-nodal graph further detailed in Section III . Figure 2: (a) The steps involved in the analysis shown in the current work are presented in a block fashion. (b) Data from Dataset 1 is transformed from spike-versus-time to a multi-nodal graph format with value of complexity metrics shown below. One crucial step in implementing such fluidic reservoir computing algorithms is that the response of the physical system to an input impulse or signal is characterised and used for training a given neural network architecture. A plenty of such architectures exist in the literature, such as the Spiking Neural Networks (SNNs) [ 9 ] , Extreme Learning Machines (ELMs) [ 6 ] , Echo State Networks (ESNs) [ 10 ] , Recurrent Neural Network (RNN) [ 11 ] , or more broadly, neuromorphic computing (NMC) architectures [ 12 ] , [ 13 ] , [ 14 ] . While each of such networks have their own dictionary and methodology required to implement them, a generic fluidic system (proteinoid gel in our case), need a fundamental “theory of computing (ToC)” framework that grants them inherent capability of universal computing, upon which a suitable NMC methodology can be implemented. Figure 3: Raw Spike Trains Datasets: The figure shows “spikes or no spikes” (0s or 1s) in the proteinoid samples prepared in Section I plotted against time (s) and recorded using voltage sensitive dyes (VSDs). In the current work, we record the electrical activity in proteinoid samples in the form of proteinoid spiking data. Five distinct datasets of the spiking trains extracted from the proteinoids are transformed using a novel modality, which is the combination of two functions: F_{1} (spiral sampling) and F_{2} (significant digit extraction). After this transformation, the complexity of the transformed data is calculated using eight graph-based discrete metrics which are then unified into a single meta-metric . Finally, the prediction algorithm for the proteinoid spikes is constructed using a feedforward neural network architecture, which captures three class of features (temporal, statistical, and spectral) comprised of a 16-dimensional vector space. The algorithm is, effectively, classification model which works with 70.41% accuracy (owing to a restricted data set from the VSD measurements). Broadly speaking, our analysis lets us motivate future directions for this research towards more universal (chemical) computing paradigms, such as proteinoid transformers or (advanced) proteinoid physical generative AI models, as opposite to primitive information-theoretic agents [ 15 ] , which can inspire experimentalists to come up with the physical embodiments of proteinoid GPTs - the collection of which can serve as the universal-computing multiverse with applications ranging from organics electronics technologies (OECTs) [ 16 ] , microfluidics & drug discovery [ 17 ] , to catalytic nanomotors [ 18 ] and beyond, and amenable for further relatively topical investigations, such as the ‘c-word’ of such physical systems. I Protocol for measuring electrical activity inside proteinoids The proteinoids are prepared using the protocol used before by us [ 19 ] and reviewed thoroughly here [ 20 ] . The experimental setup is described as the following: amino acids in powdered form; 3-hole round bottom flasks; Tri-block heater; N_{2} gas; dialysis cellophane membrane; magnetic stirrer; and a water bath. 1.5 grams each of L-Aspartic acid, L-Histidine, L-Phenylalanine, L-Glutamic acid, and L-Lysine is mixed-and-heated in the (3-hole) round bottom flask at 290 ∘ C. The temperature is step-by-step increased by 10 ∘ . After fuming start to appear, the exhaust in the fumehood (with N 2 gas as the inlet) is started, to release the fumes out. The powder goes through a colour transformation from white to green colour, followed by a morphological transition resulting in the formation of a sequence of simmering microspheres. After ceasing the heating process, the remaining material solidifies and is subsequently extracted and allowed to cool for a duration of thirty minutes. The collected residue is then placed within the Slide-A-Lyzer mini dialysis apparatus, with 10,000 molecular weight cut-off and

Read the complete text