The widespread deployment of Large Language Models (LLMs) across various sectors has underscored their versatility beyond conventional natural language processing applications. Although LLMs are adept at analyzing time series and geospatial data, their capacity to process human spatio-temporal activity data is not yet fully explored. To bridge this research gap, we introduce "LLM-Next," a model engineered to understand spatio-temporal data for predicting a user’s next Point-of-Interest (POI). We developed a specific data processing approach optimized for both temporal and spatial data and fine-tuned the LLM to improve its contextual awareness of human mobility patterns. Comparative experiments on three distinct datasets of human mobility in the real world demonstrate that LLM-Next outperforms existing baseline models in accuracy, thus redefining benchmarks in the domain.

In this paper, we explore if we can personalize horror games in Virtual Reality using auditory and visual stimuli with the help of different machine learning algorithms. Based on the heart rate of the subjects we personalize the sound and lighting effects of the game in two different environmental settings. Gradient Boosted Tree Regression, Random Forest Regression, and Tree Ensemble Regression were used to predict which sound and lighting effects should be used in subsequent levels to increase the horror aspect. In order to have a realistic game experience and due to the ongoing coronavirus pandemic, participants were recruited online. Participants could play the game wherever and whenever they wanted. The participants were also asked to complete Self-Assessment Manikin tests after playing the game. We present our discussions and observations of how different factors affect the heart rate in the game and if the heart rate data aligns with the participant’s Self-Assessment Manikin test data.

The structural class information about a protein is important to understand its biological properties. NMR is one of the most powerful tools to obtain structural information of proteins in atomic resolution. However, an analysis of protein three-dimensional structure from NMR spectra usually requires laborious chemical shift assignment. We developed a new method for predicting the protein structural class directly from the NMR spectra without any chemical shift assignment. The results show that our method outperforms the methods using current secondary structure prediction.

Protein engineering, developing novel proteins with a desired activity, has become increasingly important in many fields. This paper presents two studies in protein engineering: (i) a biological implementation of a genetic algorithm, with an observed in vitro evolution, and (ii) its preliminary computer simulation using a prototypical probabilistic model based on a random walk. The steady evolution of the fitness distribution of the mutant proteins that appeared in the biological experiments has provided some convincing evidence about the search behavior and the fitness landscape. The computer simulation and the simple probabilistic model have indicated their future potential for providing a practical alternative to the time-consuming manual operations in the biological experiments. Successful experimental results in the two studies have raised expectations of their further development and mutually beneficial interactions.

MOTIVATION: Large-scale biological analyses produce huge amounts of data. As a consequence, automation in the data analysis process is needed. Sample screening problems in NMR high-throughput protein structure analysis are the typical examples. Especially, screening by protein (1)H-(15)N heteronuclear single quantum coherence (HSQC) spectra must be done quantitatively by a human expert. One popular solution for this problem is data mining. Machine learning methods can automatically extract rules and achieve high accuracy in prediction when a good quality training dataset is prepared. However, they tend to be a black box and the learned machines suffer the risk of overfitting to the dataset. RESULTS: We propose a model which evaluates HSQC spectra for feature construction. The model calculates similarity between the measured chemical shifts and those of a random coil peak model. We applied our feature construction method for the machine learning discrimination of folded protein HSQC spectra from unfolded ones, and compared our model-based features with those of conventional sequence-based features and image recognition features. The results revealed that our method has sufficient discrimination power and less overfits on training data, as compared to the other methods. In addition, our method succeeded reduction of input data complexity towards further investigation. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Early explosion of DNA computing to solve combinatorial problems is now shrinking by three hardness; (1) code set design, (2) scalabfty and (3) speed and reliability. This paper proposes an implementation of Aqueous memory molecules by using light responsive modification of DNAs and show a series of feasibility experiments. We expect to overcome three difficulties since Aqueous computing is code design free, DNA sequence provides arbitrary size of address space, and light responsive reaction is fast and reliable.

AntNet-FA showed high potential for best-effort routing through some experiments. However this algorithm was pointed out that it can't adapt to a situation of traffic change such as congestion. Loop-free constraint is effective for sparse networks, but the constraint has a side effect for scale-free networks. In this paper, we propose an AntNet-based algorithm, AntNet-CHTT (clearing history of trip time). A key point of the algorithm is to determine the case that congestion occurs but it isn't balanced. And then this congestion is considered not to be transient, the node launches forward ants with loop-free constraint to search other routes for a destination while the congestion continues. Furthermore, the algorithm clears the history of trip time to urge the discovery of other routes, so the algorithm can use network resource effectively. We tested the algorithms on three networks. One of these networks was a sparse network like ring topology and two networks of these networks were a scale-free network.

The evolution of biological functions in protein molecules may take place on two layers that are local fragment and global domain conservations/mutations. Based on the fact that the three-dimensional structure of a protein activates its native function, this paper discusses acquiring such biological importance from comparing protein structures. Unlike the one-point search of existing methods based on various ideas, our approach utilizes the population search ability of real-coded genetic algorithm that is asynchronously parallelized. It may be useful to optimize this issue using a multiple objective evolutionary approach. In this work, we focus on maximizing two fitness functions because of the obvious trade-off. Our method as a generic structure-based alignment tool can compare all types of proteins on the two layers at a time. As the most advantageous fact, the genetic algorithm preserves local alignments as building blocks and reuses them for finding global alignments. This feature gives information on the connectivity of local fragments that often involve biological important parts, such as binding sites, active sites, etc. Robust optimization of our approach appears from experiment of protein pairs that are functionally and structurally similar/distinct. The results show that the proposed method is able to pull out the significant consideration to biological analyses.

Extracting biological importance from protein structures is extremely helpful to understand the molecular nature. Although methods for protein structure-based alignment have been hitherto proposed in a number of ways, each method focuses on a part of alignment possibility. We have developed a generic method for pair wise structure-based alignment utilizing the population search ability of a real-coded genetic algorithm. Our method simultaneously optimizes vector-expressed local fragment posture and global atomic superposition. Here, we report comparative results derived from the proposed method and existing methods. The experiments use three protein pairs well studied and a number of pairs derived from diverse protein families. The results show that our method provides useful two-layer similarity and statistical significance at a time to be able to capture not only the remarkable difference between local alignment and global alignment but also biologically meaningful common folds and motifs. Interestingly, we unveiled a vague region in protein structure-function relationships. It may indicate the limit of using alpha-carbon backbones.

MOTIVATION: The ultimate goal of bioinformatics is to reconstruct biological systems in the computer. Since biological systems have many levels, it is important to focus on an appropriate level. In our first application of computer modeling to the early development of the nematode Caenorhabditis elegans, we focus on the cellular arrangement in early embryos. This plays a very important role in cell fate determination by cell-cell interaction, and is regarded as a system, one level higher than the system of gene regulation within cells. It is largely restricted by physical conditions that seemed feasible to model by computer. RESULTS: We constructed a computer model of the C.elegans embryo, currently up to the 4-cell stage, using a deformable and dividable triangulated network. The model is based solely on cellular-level dynamics. We found that the optimal ranges of three parameters that affect the elongation of dividing cells led, in computer simulations, to almost the same cellular arrangements as in real embryos. The nature of the model and the relationship with real embryos are discussed.

We propose a novel method of self-organizing formation. It is assumed that elements are not connected to each other and they can move in continuous space. The objective is to arrange elements in a certain spatial pattern like a crystal, and to make the outline of the group in the desired shape. For this purpose, we propose a method by using virtual springs among the elements. In this algorithm, an element generates virtual springs between the neighbor element based on information of how many other elements exist in the neighborhood with a certain radius. Although the elements interact locally, only by virtual springs, and they do not have global information at all, they form a shape much larger than the sensory radius. By a simulation study, we confirmed convergence to a target shape from a random state in very high probability. This kind of algorithm gives a new principle of self-organizing formation, and its simplicity will be useful for the design of self-assembling nano machines in future.

The paper applies a method, Genetic algorithm with Search area Adaptation (GSA), to function optimization. In a previous study (H. Someya and M. Yamamura, 1999), GSA was proposed for the floorplan design problem and it showed better performance than several existing methods. We believe that investigation of the searching behavior of the algorithm is important. However, since the floorplan design problem is a combinatorial optimization problem, we do not know in detail why GSA works well. Thus, we apply GSA to function optimization in order to study the searching behavior in detail. In the function optimization, several benchmarks have been proposed, and their optima and landscapes are known. There is another reason to apply GSA to function optimization: we would like to propose a superior method for function optimization. Through several experiments, we have confirmed that GSA works adaptively and it shows higher performance than existing methods.

Molecular computing is viewed here as a process of writing on molecules while they are dissolved in water. When DNA molecules are employed, they are used only in double stranded form and only as data registers. All computations are initialized with the same single molecular variety. Current progress toward laboratory prototyping of computations is reported.