Set Of Descriptors Research Articles

Physics-informed machine learning prediction of the martensitic transformation temperature for the design of “NiTi-like” high entropy shape memory alloys

The present study proposes a physics-informed machine learning (PIML) algorithm-based approach aimed at predicting the martensitic transformation temperature (Ms) for the design of “NiTi-like” high entropy shape memory alloys (HESMAs). A previously established HESMAs database is enriched and extended to include binary, ternary, quaternary, quinary and senary alloys containing the most employed alloying elements for HEAs design such as Ni equivalents (Fe, Cu, Co, Pd, Pt and Au), Ti equivalents (Zr and Hf), Nb and Ta. The Extremely Randomized Trees algorithm, based on the concept of multiple random decision tree predictions, is adopted as the regression method for Ms temperature prediction. Two strategies for the algorithm inputs have been adopted, discussed, and compared in terms of reliable predictions. The first relies on the composition of the alloying elements, whereas the second exploits a defined set of intrinsic material descriptors. The latter are based on mixing enthalpy, atomic radius, electronegativity, atomic number and number of elements. A high accuracy of the MS prediction has been reached when considering the material descriptors. In fact, the second strategy induces a mean absolute error that is less than 30 °C for alloys containing up to 4 elements. For more elements there are more discrepancies due to the homogenization state required for HEAs. The validation of the developed approach has been performed using 6 home-made HESMAs prepared specifically for this study. It demonstrated the predictive capabilities of the developed physics-informed machine learning based approach. Finally, a HESMA design tool has been implemented to virtually design new HESMAs with a targeted Ms temperature above 400 °C. It is worth noting that this aspect is one of the most challenging engineering issues for such alloys. An illustrative case applied to the (NiCuPd)50(TiZr)50 family of alloys demonstrates the predictive capabilities of the developed approach to design such alloys to achieve a Ms temperature in the range of 300 °C to 700 °C.

Physics infused machine learning force fields for 2D materials monolayers

Large-scale atomistic simulations of two-dimensional (2D) materials rely on highly accurate and efficient force fields. Here, we present a physics-infused machine learning framework that enables the efficient development and interpretability of interatomic interaction models for 2D materials. By considering the characteristics of chemical bonds and structural topology, we have devised a set of efficient descriptors. This enables accurate force field training using a small dataset. The machine learning force fields show great success in describing the phase transformation and domain switching behaviors of monolayer Group IV monochalcogenides, e.g., GeSe and PbTe. Notably, this type of force field can be readily extended to other non-transition 2D systems, such as hexagonal boron nitride (h BN), leveraging their structural similarity. Our work provides a straightforward but accurate extension of simulation time and length scales for 2D materials.

Using Generic Direct M-SVM Model Improved by Kohonen Map and Dempster–Shafer Theory to Enhance Power Transformers Diagnostic

Many power transformers throughout the world are nearing or have gone beyond their theoretical design life. Since these important assets represent approximately 60% of the cost of the substation, monitoring their condition is necessary. Condition monitoring helps in the decision to perform timely maintenance, to replace equipment or extend its life after evaluating if it is degraded. The challenge is to prolong its residual life as much as possible. Dissolved Gas Analysis (DGA) is a well-established strategy to warn of fault onset and to monitor the transformer’s status. This paper proposes a new intelligent system based on DGA; the aim being, on the one hand, to overcome the conventional method weaknesses; and, on the other hand, to improve the transformer diagnosis efficiency by using a four-step powerful artificial intelligence method. (1) Six descriptor sets were built and then improved by the proposed feature reduction approach. Indeed, these six sets are combined and presented to a Kohonen map (KSOM), to cluster the similar descriptors. An averaging process was then applied to the grouped data, to reduce feature dimensionality and to preserve the complete information. (2) For the first time, four direct Multiclass Support Vector Machines (M-SVM) were introduced on the Generic Model basis; each one received the KSOM outputs. (3) Dempster–Shafer fusion was applied to the nine membership probabilities returned by the four M-SVM, to improve the accuracy and to support decision making. (4) An output post-processing approach was suggested to overcome the contradictory evidence problem. The achieved AUROC and sensitivity average percentages of 98.78–95.19% (p-value < 0.001), respectively, highlight the remarkable proposed system performance, bringing a new insight to DGA analysis.

A call for clarity: a scoping review of predictors of poor outcome after emergency abdominal surgery for inflammatory bowel disease.

The medical management of inflammatory bowel disease (IBD) is rapidly progressing; however, many patients with the disease still require surgery. Often this is done as an emergency. Initiatives such as the National Emergency Laparotomy Audit have shown how evidence-based emergency surgery improves outcomes for the patient. The aim of this scoping review is to describe the current evidence base on risk stratification in emergency abdominal surgery for IBD. A literature search, abstract and full paper screening resulted in 17 articles representing 63 472 patients from seven countries. It is likely that age, the American Society of Anesthesiologists grade, comorbidity and organ dysfunction play a similar role in risk stratification in IBD patients as in other emergency abdominal surgery cohorts. However, the reporting of what is considered an IBD emergency is variable. Six studies include clear definitions of emergency in our study. The range of what is considered an emergency is within 12 h of admission to any time within an unplanned admission. To have data driven, evidence-based emergency surgical practice in IBD we need consistency of reporting, including the definitions of emergency and urgency. Core descriptor sets in IBD would be valuable.

Machine learning study on organic solar cells and virtual screening of designed non-fullerene acceptors

Machine learning (ML) is effective to establish the complicated trilateral relationship among structures, properties, and photovoltaic performance, which is fundamental issue in developing novel materials for improving power conversion efficiency (PCE) of organic solar cells (OSCs). Herein, we constructed the database of 397 donor–acceptor pairs of OSCs with photovoltaic parameters and descriptor sets, which include donor–acceptor weight ratio within the active layer of the OSCs, root mean square of roughness, and 1024-bit Morgan molecular fingerprint for donor (Fp-D) and acceptor (Fp-A). The ML models random forest (RF), adaptive boosting (AdaBoost), extra trees regression, and gradient boosting regression trees were trained based on the descriptor set. The metrics determination coefficient (R2), Pearson correlation coefficient (r), root mean square error, and mean absolute error were selected to evaluate ML model performances. The results showed that the RF model exhibits the highest accuracy and stability for PCE prediction among these four ML models. Moreover, based on the decomposition of non-fullerene acceptors L8-BO, BTP-ec9, AQx-2, and IEICO, 20 acceptor molecules with symmetric A–D–A and A–π–D–π–A architectures were designed. The photovoltaic parameters of the designed acceptors were predicted using the trained RF model, and the virtual screening of designed acceptors was conducted based on the predicted PCE. The results indicate that six designed acceptors can reach the predicted PCE higher than 12% when P3HT was adopted as a donor. While PM6 was applied as a donor, five designed acceptors can achieve the predicted PCE higher than 16%.

Cometary dust collected by MIDAS on board Rosetta

Context. The MIDAS (Micro-Imaging Dust Analysis System) atomic force microscope on board the Rosetta comet orbiter investigated and measured the 3D topography of a few hundred of nm to tens of μm sized dust particles of 67P/Churyumov-Gerasimenko with resolutions down to a few nanometers, giving insights into the physical processes of our early Solar System. Aims. We analyze the shapes of the cometary dust particles collected by MIDAS on the basis of a recently updated particle catalog with the aim to determine which structural properties remained pristine. Methods. We develop a set of shape descriptors and metrics such as aspect ratio, elongation, circularity, convexity, and particle surface and volume distribution, which can be used to describe the distribution of particle shapes. Furthermore, we compare the structure of the MIDAS dust particles and the clusters in which the particles were deposited to those found in previous laboratory experiments and by Rosetta/COSIMA. Finally, we combine our findings to calculate a pristineness score for MIDAS particles and determine the most pristine particles and their properties. Results. We find that the morphological properties of all cometary dust particles at the micrometer scale are surprisingly homogeneous despite originating from diverse cometary environments (e.g., different collection targets that are associated with cometary activities/source regions and collection velocities/periods). There is only a weak trend between shape descriptors and particle characteristics such as size, collection targets, and cluster morphology. We next find that the types of clusters found by MIDAS show good agreement with those defined by previous laboratory experiments, however, there are some differences to those found by Rosetta/COSIMA. Furthermore, our pristineness score shows that almost half of MIDAS particles suffered severe alteration by impact, which indicates structural modification by impact (e.g., flattening and/or fragmentation) is inevitable despite the very low collection speeds (i.e., ~3–7 m s−1). Based on our result, we rate 19 out of 1082 MIDAS particles at least moderately pristine that is they are not substantially flattened by impact, not fragmented, and/or not part of a fragmentation cluster.

Morphology and Transport Characterization of Catalyst Layers for CO2 Reduction

Gas diffusion electrodes (GDEs) help to reduce transport limitations in devices for electrochemical CO2 reduction. Homogenized modeling of such devices requires input of morphological characteristics and effective transport properties of the porous structure, which can be obtained by pore-scale methods. Due to the small pore sizes and layer thicknesses, such characterization of the catalyst layer (CL) is difficult. In this work, CL structures digitalized by FIB-SEM nano-tomography were analyzed to obtain a set of morphological descriptors and were considered as geometrical domains in direct pore-scale simulations to calculate tensors of effective diffusion coefficient, permeability, tortuosity, and effective ionic and electronic conductivity. The CL properties vary significantly depending on the diffusion and flow direction, and samples of similar composition and synthesis technique exhibit different transport behaviors. The reported results provide a quantitative morphological analysis and a reliable set of effective properties to be used in homogenized device modeling, a net improvement over the most commonly used empirical or analytical formulations. The data set was obtained directly from the pore-scale structure of real CLs and contributes to making device-scale homogenized modeling of GDEs more accurate and reliable.

Diagnosing Tropical Cyclone Intensity Variations from the Surface Wind Field Evolution

Abstract Tropical cyclone (TC) intensity fluctuations remain a challenge for TC forecasters. Occurring through a wide range of processes, such as vortex contraction, eyewall replacements, or emission of vortex Rossby waves, they are inherently multiscale, transient, and asymmetric. In a recent study, estimates of surface wind field inner-core properties from high-resolution satellite observations were spotted as valuable for the improvement of intensity variations statistical predictability. The present study evaluates how the temporal evolution of the vortex structure, at scales ranging from O(1) km to vortexwide, further provides insights on the modulation of intensity. The study is based on a set of seven realistic TC simulations with 1-km grid spacing. The surface wind field structure is studied through an original set of descriptors that characterize the radial profile, the azimuthal asymmetries, and their spectral distribution. While radial gradients evolve concurrently with intensity, the azimuthal variability of the inner core shows a stronger connection with shorter-scale intensity modulation. The increase of high-wavenumber asymmetries distributed around the ring of maximum winds is shown to precede phases of rapid (re)intensification by 5–6 h, while the concentration of asymmetry in wavenumbers 1 and 2 leads to intensity weakening. A machine learning classification finally highlights that the classification of intensification phases (i.e., intensification or weakening) can be improved by at least 11% (thus reaching ∼75%) when accounting for the evolution of the radial wind gradient and the variance distribution among scales in the ring of maximum wind, relative to the sole use of vortex-averaged parameters. Significance Statement The purpose of this study is to relate changes in the surface wind field structure of tropical cyclones to their intensity variations. We design an original set of parameters to characterize the inner- and near-core contraction and asymmetry, and evaluate their connection with TC intensity modulations in a set of realistic high-resolution numerical simulations. The main outcome of our study is that opposite trends in high- and low-wavenumber variance tend to occur prior to intensity changes on short time scales, with more widespread and local-scale asymmetry corresponding to intensification, and vortex-scale polarized asymmetry corresponding to weakening. These diagnoses are shown to improve the classification of intensification and weakening phases. These results advocate for enhancing real-time high-resolution observations of surface wind fields under TCs, and using asymmetry distribution as predictors in statistical forecast models.

Accelerated screening of sensitive and selective MoO3-based gas sensing materials by combining first-principles and machine learning approach

A large amount of hybrid metal oxides can be proposed for detecting various gases from rapidly developing of 2D materials and elemental doping technologies, and the key issue is how to search the large parameter space in an efficient way. First-principles approach can calculate molecular-scale electronic properties and provide a guide of material design for experiments, however, it is too time-consuming to explore all the possible metal oxides doped with different elements. Herein, we develop a novel framework via combining first-principles and machine learning, with the aim of enabling more efficient and practical screening of MoO3-based gas sensors than by experiments or first-principles alone. Owing to the high accuracy of first-principles calculations verified by experimental results, we demonstrate the reliability of the proposed methods for evaluating gas sensing performance. By proposing a set of new descriptors including d-band center and average bond length with demanding low-cost calculations, we significantly reduce the amount of required training data. In particular, the gradient boosting regression algorithm exhibits highR-square value of 0.96 and low mean absolute error value of 0.22, indicating superior reliability of the model. This work opens an avenue for quickly screening of novel metal oxides and other nano-materials with superior sensitivity and selectivity for gas detection.

Reactivity Prediction of Cu-Catalyzed Halogen Atom Transfer Reactions Using Data-Driven Techniques.

In catalysis, linear free energy relationships (LFERs) are commonly used to identify reaction descriptors that enable the prediction of outcomes and the design of more effective catalysts. Herein, LFERs are established for the reductive cleavage of the C(sp3)-X bond in alkyl halides (RX) by Cu complexes. This reaction represents the activation step in atom transfer radical polymerization and atom transfer radical addition/cyclization. The values of the activation rate constant, kact, for 107 Cu complex/RX couples in 5 different solvents spanning over 13 orders of magnitude were effectively interpolated by the equation: log kact = sC(I + C + S), where I, C, and S are, respectively, the initiator, catalyst, and solvent parameters, and sC is the catalyst-specific sensitivity parameter. Furthermore, each of these parameters was correlated to relevant descriptors, which included the bond dissociation free energy of RX and its Tolman cone angle θ, the electron affinity of X, the radical stabilization energy, the standard reduction potential of the Cu complex, the polarizability parameter π* of the solvent, and the distortion energy of the complex in its transition state. This set of descriptors establishes the fundamental properties of Cu complexes and RX that determine their reactivity and that need to be considered when designing novel systems for atom transfer radical reactions. Finally, a multivariate linear regression (MLR) approach was adopted to develop an objective model that surpassed the predictive capability of the LFER equation. Thus, the MLR model was employed to predict kact values for >2000 Cu complex/RX pairs.

How to identify pollen like a palynologist: A prior knowledge-guided deep feature learning for real-world pollen classification

Airborne pollen identification is crucial to help patients prevent pollinosis symptoms. Existing data-driven methods rely on large-scale pollen images with simple backgrounds. In real scenarios, the background is complex and the data scale is small. Therefore, these methods suffer from two challenges: (1) Irrelevant information interference; (2) Incomplete feature attention. To overcome these challenges, we propose a prior knowledge-guided deep feature learning (PK-DFL) for real-world optical microscope image classification. Its main steps are as follows: Pollen location is designed to locate pollen grains based on color features, aiming to boost the accuracy of shape and texture prior feature extraction. Shape-texture awareness helps to extract the shape and texture of pollen grains via predefined feature extractors (i.e., a set of shape descriptors and an improved SFTA). These features are used to construct two types of prior knowledge, namely shape-texture attention maps (STA maps) and shape-texture feature vectors (STF vectors). Pollen classification uses a deep network (CNN) to classify pollen via imitating the pollen identification procedure of palynologists. It uses STA maps to weight pollen images and convolutional feature maps for instructing the CNN to focus on critical areas of pollen images (for the first challenge). STF vectors are employed to obtain the inter-class similarity of pollen via template matching. This information is further converted to soft targets that are used to supervise the CNN attending to comprehensive key features (for the second challenge). Extensive experiments on real-world datasets demonstrate the effectiveness of our PK-DFL (with accuracy and F1-score over 88%).

Cell-penetrating peptides predictors: A comparative analysis of methods and datasets.

Cell-Penetrating Peptides (CPP) are emerging as an alternative to small-molecule drugs to expand the range of biomolecules that can be targeted for therapeutic purposes. Due to the importance of identifying and designing new CPP, a great variety of predictors have been developed to achieve these goals. To establish a ranking for these predictors, a couple of recent studies compared their performances on specific datasets, yet their conclusions cannot determine if the ranking obtained is due to the model, the set of descriptors or the datasets used to test the predictors. We present a systematic study of the influence of the peptide sequence's similarity of the datasets on the predictors' performance. The analysis reveals that the datasets used for training have a stronger influence on the predictors performance than the model or descriptors employed. We show that datasets with low sequence similarity between the positive and negative examples can be easily separated, and the tested classifiers showed good performance on them. On the other hand, a dataset with high sequence similarity between CPP and non-CPP will be a hard dataset, and it should be the one to be used for assessing the performance of new predictors.

RECONOCIMIENTO AUTOMÁTICO DE AUTOMÓVILES MEDIANTE MACHINE LEARNING

In this work, we perform the automatic classification of 1,000 images of five different models of automobiles. To obtain the highest precision, we have used two different classification scenarios, three algorithms, and five metrics. Also, we assume that the results can be improved by extracting the image characteristics using descriptors and using them as input. Then, we used two descriptors: a histogram of oriented gradient and a convolutional neural network ResNet-50. Our results show that the descriptors improve the classification results and obtain the highest value for the accuracy metric of 88.01 % using the ResNet-50 as a descriptor, the Training and Test Set as a scenario, and Vector Support Machine as the classification algorithm. Keywords: Convolutional Neural Networks, Gradient Oriented Histogram, Machine Learning, Fine Grain Classification, Car Images.

Descriptor-Free Deep Learning QSAR Model for the Fraction Unbound in Human Plasma.

Chemical-specific parameters are either measured in vitro or estimated using quantitative structure-activity relationship (QSAR) models. The existing body of QSAR work relies on extracting a set of descriptors or fingerprints, subset selection, and training a machine learning model. In this work, we used a state-of-the-art natural language processing model, Bidirectional Encoder Representations from Transformers, which allowed us to circumvent the need for calculation of these chemical descriptors. In this approach, simplified molecular-input line-entry system (SMILES) strings were embedded in a high-dimensional space using a two-stage training approach. The model was first pre-trained on a masked SMILES token task and then fine-tuned on a QSAR prediction task. The pre-training task learned meaningful high-dimensional embeddings based upon the relationships between the chemical tokens in the SMILES strings derived from the "in-stock" portion of the ZINC 15 dataset─a large dataset of commercially available chemicals. The fine-tuning task then perturbed the pre-trained embeddings to facilitate prediction of a specific QSAR endpoint of interest. The power of this model stems from the ability to reuse the pre-trained model for multiple different fine-tuning tasks, reducing the computational burden of developing multiple models for different endpoints. We used our framework to develop a predictive model for fraction unbound in human plasma (fu,p). This approach is flexible, requires minimum domain expertise, and can be generalized for other parameters of interest for rapid and accurate estimation of absorption, distribution, metabolism, excretion, and toxicity.

Enhancing the Search Performance of Bayesian Optimization by Creating Different Descriptor Datasets Using Density Functional Theory.

Descriptors calculated from molecular structure information can be used as explanatory variables in Bayesian optimization (BO). Even though structural and descriptor information can be obtained from various databases for general compounds, information on highly confidential compounds such as pharmaceutical intermediates and active pharmaceutical ingredients cannot be retrieved from these databases. In particular, determining the stable structure and electronic state of a compound via quantum chemical calculations from descriptor information requires considerable computational time. Although descriptor information can be obtained using density functional theory (DFT), which has a relatively light computational load, only conventional combinations of basis sets and functionals can be selected before experiments instead of the best ones. Few studies have discussed these effects on the search performance of BO, and good search performance is highly dependent on the application. Therefore, we developed a method to improve the search performance of BO by using descriptors computed from several combinations of basis sets and functionals. The dataset obtained from averaging multiple descriptor sets exhibited better BO search performance than that of a single descriptor dataset. In addition, the more descriptor sets used for averaging, the better the search performance. This method has a relatively small computational load and can be easily used by those who are unfamiliar with quantum chemical calculations.

A quantum chemical prediction on arc interruption capability of the dielectric gases

AbstractThe use of sulfur hexafluoride (SF6) as an electrical insulator and arc‐quenching gas in high‐voltage equipment raises serious environmental concerns. The search of a replacement gas for SF6 is hindered by a priori assessment on its interruption performance. The routine fluid or mathematic magneto‐hydrodynamic models for arc burning and extinguishing are too complex to be practical for virtual screening. Herein a state‐of‐the‐art quantum chemical model to predict the interruption capability of the dielectric gases straightforwardly has been established in terms of the rate of rise of the recovery voltage (RRRV). On the basis of the molecular electrostatic potentials, five sets of descriptors, including the global statistical parameters and , the site‐specific parameter ΔVs, the total positive surface area As+, augmented with the product of polarizability and dipole moment αμ, were optimized to reveal the inherent mechanisms for interruption and thus a viable structure–activity relationship mode for RRRV has been developed with the correlation coefficient 0.975. Theoretical RRRV relative to SF6 = 100 for all the known dielectric gases are in good agreement with the experimental data by a mean absolute deviation of 3.6. Accordingly, the perfluorinated cycloalkenes and alkynes, in particular, 1,3,3,3‐tetrafluoropropyne, are found to be the promising candidates as the replacement dielectric gases for SF6.

SP5.11 A systematic review of participant descriptors reported in studies of small bowel obstruction

Abstract Aim Reporting of participant descriptors in studies of small bowel obstruction (SBO) can help identify characteristics associated with favourable outcomes and allow comparison with other studies and real-world clinical populations. The aim was to identify the pattern of participant descriptors reported in studies assessing interventions for SBO. Method This systematic review was registered with PROSPERO (CRD42021281031) and reported in line with the PRISMA checklist. Systematic searches of Ovid MEDLINE, Embase, and Cochrane Central Register of Controlled Trials (CENTRAL) were undertaken to identify studies assessing operative and non-operative interventions for adults with SBO. Studies were dual screened for inclusion. Descriptors were categorised into conceptual domains by the research team. Results Searches identified 2648 studies, of which 85 were included. A total of 168 unique descriptors were identified. On average, studies reported 12 descriptors. The most frequently reported descriptors were sex, age, SBO aetiology, history of abdominal surgery, BMI and ASA classification. The highest number of descriptors in a single study was 34, compared to the lowest number of descriptors which was one. Pathway factors were the least frequently described domain. Overall, 43 descriptors were reported in just one study. Conclusion There is a lack of consistency in participant descriptors reported in studies of SBO. Furthermore, a significant proportion of the descriptors were used infrequently. This makes it challenging to assess whether study participants are representative of the wider population. Further work is required to develop a Core Descriptor Set to standardise the reporting of patient characteristics and reduce heterogeneity between studies.

General structure-activity/selectivity relationship patterns for the inhibitors of the chemokine receptors (CCR1/CCR2/CCR4/CCR5) with application for virtual screening of PubChem database

CC chemokine receptors (CCRs) form a crucial subfamily of G protein-linked receptors that play a distinct role in the onset and progression of various life-threatening diseases. The main aim of this research is to derive general structure-activity relationship (SAR) patterns to describe the selectivity and activity of CCR inhibitors. To this end, a total of 7332 molecules related to the inhibition of CCR1, CCR2, CCR4, and CCR5 were collected from the Binding Database and analyzed using machine learning techniques. A diverse set of 450 molecular descriptors was calculated for each molecule, and the molecules were classified based on their therapeutic targets and activities. The variable importance in the projection (VIP) approach was used to select discriminatory molecular features, and classification models were developed using supervised Kohonen networks (SKN) and counter-propagation artificial neural networks (CPANN). The reliability and predictability of the models were estimated using 10-fold cross-validation, an external validation set, and an applicability domain approach. We were able to identify different sets of molecular descriptors for discriminating between active and inactive molecules and model the selectivity of inhibitors towards different CCRs. The sensitivities of the predictions for the external test set for the SKN models ranged from 0.827–0.873. Finally, the developed classification models were used to screen approximately 2 million random molecules from the PubChem database, with average values for areas under the receiver operating characteristic curves ranging from 0.78–0.96 for SKN models and 0.75–0.89 for CPANN models. Communicated by Ramaswamy H. Sarma

Underwater object classification combining SAS and transferred optical-to-SAS Imagery

Combining synthetic aperture sonar (SAS) imagery with optical images for underwater object classification has the potential to overcome challenges such as water clarity, the stability of the optical image analysis platform, and strong reflections from the seabed for sonar-based classification. In this work, we propose this type of multi-modal combination to discriminate between man-made targets and objects such as rocks or litter. We offer a novel classification algorithm that overcomes the problem of intensity and object formation differences between the two modalities. To this end, we develop a novel set of geometrical shape descriptors that takes into account the geometrical relation between the object’s shadow and highlight. Results from 7,052 pairs of SAS and optical images collected during several sea experiments show improved classification performance compared to the state-of-the-art for better discrimination between different types of underwater objects. For reproducability, we share our database.

Genetic descriptor search algorithm for predicting hydrogen adsorption free energy of 2D material

Transition metal dichalcogenides (TMDs) have emerged as a promising alternative to noble metals in the field of electrocatalysts for the hydrogen evolution reaction. However, previous attempts using machine learning to predict TMD properties, such as catalytic activity, have been shown to have limitations in their dependence on large amounts of training data and massive computations. Herein, we propose a genetic descriptor search that efficiently identifies a set of descriptors through a genetic algorithm, without requiring intensive calculations. We conducted both quantitative and qualitative experiments on a total of 70 TMDs to predict hydrogen adsorption free energy (Delta G_H) with the generated descriptors. The results demonstrate that the proposed method significantly outperformed the feature extraction methods that are currently widely used in machine learning applications.