Artificial intelligence-enabled predictive energy saving planning of liquid cooling system for data centers

Advanced Engineering Informatics, Год журнала: 2025, Номер 65, С. 103283 - 103283

Опубликована: Март 31, 2025

Язык: Английский

---

From genotype to phenotype: computational approaches for inferring microbial traits relevant to the food industry

FEMS Microbiology Reviews, Год журнала: 2023, Номер 47(4)

Опубликована: Июнь 7, 2023

Abstract When selecting microbial strains for the production of fermented foods, various phenotypes need to be taken into account achieve target product characteristics, such as biosafety, flavor, texture, and health-promoting effects. Through continuous advances in sequencing technologies, whole-genome sequences increasing quality can now obtained both cheaper faster, which increases relevance genome-based characterization phenotypes. Prediction from genome makes it possible quickly screen large strain collections silico identify candidates with desirable traits. Several relevant foods predicted using knowledge-based approaches, leveraging our existing understanding genetic molecular mechanisms underlying those In absence this knowledge, data-driven approaches applied estimate genotype–phenotype relationships based on experimental datasets. Here, we review computational methods that implement knowledge- phenotype prediction, well combine elements approaches. Furthermore, provide examples how these have been industrial biotechnology, special focus food industry.

Язык: Английский

---

A Novel Transformer-CNN Approach for Predicting Soil Properties from LUCAS Vis-NIR Spectral Data

Agronomy, Год журнала: 2024, Номер 14(9), С. 1998 - 1998

Опубликована: Сен. 2, 2024

Soil, a non-renewable resource, requires continuous monitoring to prevent degradation and support sustainable agriculture. Visible-near-infrared (Vis-NIR) spectroscopy is rapid cost-effective method for predicting soil properties. While traditional machine learning methods are commonly used modeling Vis-NIR spectral data, large datasets may benefit more from advanced deep techniques. In this study, based on the library LUCAS, we aimed enhance regression model performance in property estimation by combining Transformer convolutional neural network (CNN) techniques predict 11 properties (clay, silt, pH CaCl2, H2O, CEC, OC, CaCO3, N, P, K). The Transformer-CNN accurately predicted most properties, outperforming other (partial least squares (PLSR), random forest (RFR), vector (SVR), Long Short-Term Memory (LSTM), ResNet18) with 10–24 percentage point improvement coefficient of determination (R2). excelled N (R2 = 0.94–0.96, RPD > 3) performed well clay, sand, K 0.77–0.85, 2 < 3). This study demonstrates potential enhancing prediction, although future work should aim optimize computational efficiency explore wider range applications ensure its utility different agricultural settings.

Язык: Английский

---

Deep learning probability flows and entropy production rates in active matter

Proceedings of the National Academy of Sciences, Год журнала: 2024, Номер 121(25)

Опубликована: Июнь 11, 2024

Active matter systems, from self-propelled colloids to motile bacteria, are characterized by the conversion of free energy into useful work at microscopic scale. They involve physics beyond reach equilibrium statistical mechanics, and a persistent challenge has been understand nature their nonequilibrium states. The entropy production rate probability current provide quantitative ways do so measuring breakdown time-reversal symmetry. Yet, efficient computation remained elusive, as they depend on system’s unknown high-dimensional density. Here, building upon recent advances in generative modeling, we develop deep learning framework estimate score this We show that score, together with equations motion, gives access rate, current, decomposition local contributions individual particles. To represent introduce spatially transformer network architecture learns high-order interactions between particles while respecting underlying permutation demonstrate broad utility scalability method applying it several systems active undergoing motility-induced phase separation (MIPS). single trained system 4,096 one packing fraction can generalize other regions diagram, including many 32,768 use observation quantify spatial structure departure MIPS function number fraction.

Язык: Английский

---

Transformer technology in molecular science

Wiley Interdisciplinary Reviews Computational Molecular Science, Год журнала: 2024, Номер 14(4)

Опубликована: Июль 1, 2024

Abstract A transformer is the foundational architecture behind large language models designed to handle sequential data by using mechanisms of self‐attention weigh importance different elements, enabling efficient processing and understanding complex patterns. Recently, transformer‐based have become some most popular powerful deep learning (DL) algorithms in molecular science, owing their distinctive architectural characteristics proficiency handling intricate data. These leverage capacity architectures capture hierarchical dependencies within As applications transformers science are very widespread, this review, we only focus on technical aspects technology molecule domain. Specifically, will provide an in‐depth investigation into machine techniques science. The under consideration include generative pre‐trained (GPT), bidirectional auto‐regressive (BART), encoder representations from (BERT), graph transformer, transformer‐XL, text‐to‐text transfer vision (ViT), detection (DETR), conformer, contrastive language‐image pre‐training (CLIP), sparse transformers, mobile transformers. By examining inner workings these models, aim elucidate how innovations contribute effectiveness We also discuss promising trends context emphasizing capabilities potential for interdisciplinary research. This review seeks a comprehensive that driving advancements article categorized under: Data Science > Chemoinformatics Artificial Intelligence/Machine Learning

Язык: Английский

---

Transformers and genome language models

Nature Machine Intelligence, Год журнала: 2025, Номер unknown

Опубликована: Март 13, 2025

---

Assessing the Strengths and Weaknesses of Large Language Models

Journal of Logic Language and Information, Год журнала: 2023, Номер 33(1), С. 9 - 20

Опубликована: Ноя. 11, 2023

Abstract The transformers that drive chatbots and other AI systems constitute large language models (LLMs). These are currently the focus of a lively discussion in both scientific literature popular media. This ranges from hyperbolic claims attribute general intelligence sentience to LLMs, skeptical view these devices no more than “stochastic parrots”. I present an overview some weak arguments have been presented against consider several compelling criticisms devices. former significantly underestimate capacity achieve subtle inductive inferences required for high levels performance on complex, cognitively significant tasks. In instances, misconstrue nature deep learning. latter identify limitations way which learn represent patterns data. They also point out important differences between procedures through neural networks humans acquire knowledge natural language. It is necessary look carefully at sets order balanced assessment potential LLMs.

Язык: Английский

---

PepCNN deep learning tool for predicting peptide binding residues in proteins using sequence, structural, and language model features

Scientific Reports, Год журнала: 2023, Номер 13(1)

Опубликована: Ноя. 28, 2023

Abstract Protein–peptide interactions play a crucial role in various cellular processes and are implicated abnormal behaviors leading to diseases such as cancer. Therefore, understanding these is vital for both functional genomics drug discovery efforts. Despite significant increase the availability of protein–peptide complexes, experimental methods studying remain laborious, time-consuming, expensive. Computational offer complementary approach but often fall short terms prediction accuracy. To address challenges, we introduce PepCNN, deep learning-based model that incorporates structural sequence-based information from primary protein sequences. By utilizing combination half-sphere exposure, position specific scoring matrices multiple-sequence alignment tool, embedding pre-trained language model, PepCNN outperforms state-of-the-art specificity, precision, AUC. The software datasets publicly available at https://github.com/abelavit/PepCNN.git .

Язык: Английский

---

Extracellular vesicles for developing targeted hearing loss therapy

Journal of Controlled Release, Год журнала: 2024, Номер 366, С. 460 - 478

Опубликована: Янв. 11, 2024

Язык: Английский

---

Vaccine development using artificial intelligence and machine learning: A review

International Journal of Biological Macromolecules, Год журнала: 2024, Номер unknown, С. 136643 - 136643

Опубликована: Окт. 1, 2024

Язык: Английский

---