Median Relative Error Research Articles

The application of machine learning in tidal evolution simulation of star–planet systems

ABSTRACT With the release of a large amount of astronomical data, an increasing number of close-in hot Jupiters have been discovered. Calculating their evolutionary curves using star–planet interaction models presents a challenge. To expedite the generation of evolutionary curves for these close-in hot Jupiter systems, we utilized tidal interaction models established on mesa to create 15 745 samples of star–planet systems and 7500 samples of stars. Additionally, we employed a neural network (Multilayer Perceptron – MLP) to predict the evolutionary curves of the systems, including stellar effective temperature, radius, stellar rotation period, and planetary orbital period. The median relative errors of the predicted evolutionary curves were found to be 0.15 per cent, 0.43 per cent, 2.61 per cent, and 0.57 per cent, respectively. Furthermore, the speed at which we generate evolutionary curves exceeds that of model-generated curves by more than four orders of magnitude. We also extracted features of planetary migration states and utilized lightgbm to classify the samples into six categories for prediction. We found that by combining three types that undergo long-term double synchronization into one label, the classifier effectively recognized these features. Apart from systems experiencing long-term double synchronization, the median relative errors of the predicted evolutionary curves were all below 4 per cent. Our work provides an efficient method to save significant computational resources and time with minimal loss in accuracy. This research also lays the foundation for analysing the evolutionary characteristics of systems under different migration states, aiding in the understanding of the underlying physical mechanisms of such systems. Finally, to a large extent, our approach could replace the calculations of theoretical models.

Accuracy of Automatic Processing of Speech-Language Pathologist and Child Talk During School-Based Therapy Sessions.

This study examines the accuracy of Interaction Detection in Early Childhood Settings (IDEAS), a program that automatically transcribes audio files and estimates linguistic units relevant to speech-language therapy, including part-of-speech units that represent features of language complexity, such as adjectives and coordinating conjunctions. Forty-five video-recorded speech-language therapy sessions involving 27 speech-language pathologists (SLPs) and 56 children were used. The F measure determines the accuracy of IDEAS diarization (i.e., speech segmentation and speaker classification). Two additional evaluation metrics, namely, median absolute relative error and correlation, indicate the accuracy of IDEAS for the estimation of linguistic units as compared with two conditions, namely, Oracle (manual diarization) and Voice Type Classifier (existing diarizer with acceptable accuracy). The high F measure for SLP talk data suggests high accuracy of IDEAS diarization for SLP talk but less so for child talk. These differences are reflected in the accuracy of IDEAS linguistic unit estimates. IDEAS median absolute relative error and correlation values for nine of the 10 SLP linguistic unit estimates meet the accuracy criteria, but none of the child linguistic unit estimates meet these criteria. The type of linguistic units also affects IDEAS accuracy. IDEAS was tailored to educational settings to automatically convert audio recordings into text and to provide linguistic unit estimates in speech-language therapy sessions and classroom settings. Although not perfect, IDEAS is reliable in automatically capturing and returning linguistic units, especially in SLP talk, that are relevant in research and practice. The tool offers a way to automatically measure SLP talk in clinical settings, which will support research seeking to understand how SLP talk influences children's language growth.

Large-scale prediction of collision cross-section with very deep graph convolutional network for small molecule identification

Ion mobility spectrometry (IMS) is a promising analytical technique for mass spectrometry (MS)-based compound identification by providing collision cross-section (CCS) value as an additional dimension with structural information. Here, GraphCCS was proposed to accurately predict the CCS value and expand the coverage of CCS libraries. A new adduct encoding method was proposed to encode SMILES strings and adduct types of compounds into adduct graphs. GraphCCS extended its predictive capability to ten different adduct types. A very deep graph convolutional network with up to 40 GCN layers was built to predict CCS values from adduct graphs. A curated dataset with 12,775 experimental CCS values was used to train, validate, and test the GraphCCS model. The resulting CCS predictions achieved a median relative error (MedRE) of 0.94 % and a coefficient of determination (R2) of 0.994 on the test set. Results on external test sets showed that GraphCCS outperformed AllCCS2, CCSbase, SigmaCCS, and DeepCCS. Based on the developed GraphCCS method, a large-scale in-silico database was built, including 2,394,468 CCS values. Those CCS values can be used to filter false positives complementary to retention times and tandem mass spectra. Finally, the effectiveness of GraphCCS in assisting compound identification was tested on a mouse adrenal gland lipid dataset with 1,960 lipids. The results demonstrated that the in-silico CCS values combined with MS spectra and retention times can efficiently filter the false positive candidates.

General Method for Fitting Kinetics from the SECM Images of Reactive Sites on Flat Surfaces.

Scanning electrochemical microscopy (SECM) is a technique for imaging electrochemical reactions at a surface. The interaction between electrochemical reactions occurring at the sample and scanning electrode tip is quite complicated and requires computer modeling to obtain quantitative information from SECM images. Often, existing computer models must be modified, or a new model must be created from scratch to fit kinetic parameters for different reactive features. This work presents a method that can simulate the SECM image of a reactive feature of any shape on a flat surface which is coupled to a computer program which effectuates the automated fitting of kinetic information from these images. This fitting program is evaluated along with several methods for estimating the shapes of reactive features from their SECM images. Estimates of the reactive feature shape from SECM images were not sufficiently accurate and produced median relative errors for the surface rate constant that were >50%. Fortunately, more precise techniques for imaging the reactive features such as optical microscopy can supply sufficiently accurate shapes for the fitting procedure to produce accurate results. Fits of simulated SECM images using the actual shape from the simulation produced median relative errors for the surface rate constant that were <10% for the smallest reactive features tested. This method was applied to the SECM images of aluminum alloy AA7075 which revealed diffusion-limited kinetics for ferrocene methanol reduction over inclusions in the surface of the alloy.

Predicting Collision Cross-Section Values for Small Molecules through Chemical Class-Based Multimodal Graph Attention Network.

Libraries of collision cross-section (CCS) values have the potential to facilitate compound identification in metabolomics. Although computational methods provide an opportunity to increase library size rapidly, accurate prediction of CCS values remains challenging due to the structural diversity of small molecules. Here, we developed a machine learning (ML) model that integrates graph attention networks and multimodal molecular representations to predict CCS values on the basis of chemical class. Our approach, referred to as MGAT-CCS, had superior performance in comparison to other ML models in CCS prediction. MGAT-CCS achieved a median relative error of 0.47%/1.14% (positive/negative mode) and 1.40%/1.63% (positive/negative mode) for lipids and metabolites, respectively. When MGAT-CCS was applied to real-world metabolomics data, it reduced the number of false metabolite candidates by roughly 25% across multiple sample types ranging from plasma and urine to cells. To facilitate its application, we developed a user-friendly stand-alone web server for MGAT-CCS that is freely available at https://mgat-ccs-web.onrender.com. This work represents a step forward in predicting CCS values and can potentially facilitate the identification of small molecules when using ion mobility spectrometry coupled with mass spectrometry.

Evaluating the accuracy of automated processing of child and adult language production in preschool classrooms.

Young children's language and social development is influenced by the linguistic environment of their classrooms, including their interactions with teachers and peers. Measurement of the classroom linguistic environment typically relies on observational methods, often providing limited 'snapshots' of children's interactions, from which broad generalizations are made. Recent technological advances, including artificial intelligence, provide opportunities to capture children's interactions using continuous recordings representing much longer durations of time. The goal of the present study was to evaluate the accuracy of the Interaction Detection in Early Childhood Settings (IDEAS) system on 13 automated indices of language output using recordings collected from 19 children and three teachers over two weeks in an urban preschool classroom. The accuracy of language outputs processed via IDEAS were compared to ground truth via linear correlations and median absolute relative error. Findings indicate high correlations between IDEAS and ground truth data on measures of teacher and child speech, and relatively low error rates on the majority of IDEAS language output measures. Study findings indicate that IDEAS may provide a useful measurement tool for advancing knowledge about children's classroom experiences and their role in shaping development.

Continuous Glucose Monitoring and Recreational Scuba Diving in Type 1 Diabetes: Head-to-Head Comparison Between Free Style Libre 3 and Dexcom G7 Performance.

Background: Scuba diving was previously excluded because of hypoglycemic risks for patients with type 1 diabetes mellitus(T1DM). Specific eligibility criteria and a safety protocol have been defined, whereas continuous glucose monitoring (CGM) systems have enhanced diabetes management. This study aims to assess the feasibility and accuracy of CGM Dexcom G7 and Free Style Libre 3 in a setting of repetitive scuba diving in T1DM, exploring the possibility of nonadjunctive use. Material and Methods: The study was conducted during an event of Diabete Sommerso® association in 2023. Participants followed a safety protocol, with capillary glucose as reference standard (Beurer GL50Evo). Sensors' accuracy was evaluated through median and mean absolute relative difference (MeARD, MARD) and surveillance error grid (SEG). Data distribution and correlation were estimated by Spearman test and Bland-Altman plots. The ability of sensors to identify hypoglycemia was assessed by contingency tables. Results: Data from 202 dives of 13 patients were collected. The overall MARD was 31% (Dexcom G7) and 14.2% (Free Style Libre 3) and MeARD was 19.7% and 11.6%, respectively. Free Style Libre 3 exhibited better accuracy in normoglycemic and hyperglycemic ranges. SEG analysis showed 82.1% (Dexcom G7) and 97.4% (Free Style Libre 3) data on no-risk zone. Free Style Libre 3 better performed on hypoglycemia identification (diagnostic odds ratio of 254.10 vs. 58.95). Neither of the sensors reached the MARD for nonadjunctive use. Conclusions: The study reveals Free Style Libre 3 superior accuracy compared with Dexcom G7 in a setting of repetitive scuba diving in T1DM, except for hypoglycemic range. Both sensors fail to achieve accuracy for nonadjunctive use. Capillary tests remain crucial for safe dive planning, and sensor data should be interpreted cautiously. We suggest exploring additional factors potentially influencing sensor performance.

Software cost estimation predication using a convolutional neural network and particle swarm optimization algorithm

Over the past decades, the software industry has expanded to include all industries. Since stakeholders tend to use it to get their work done, software houses seek to estimate the cost of the software, which includes calculating the effort, time, and resources required. Although many researchers have worked to estimate it, the prediction accuracy results are still inaccurate and unstable. Estimating it requires a lot of effort. Therefore, there is an urgent need for modern techniques that contribute to cost estimation. This paper seeks to present a model based on deep learning and machine learning techniques by combining convolutional neural networks (CNN) and the particle swarm algorithm (PSO) in the context of time series forecasting, which enables feature extraction and automatic tuning of hyperparameters, which reduces the manual effort of selecting parameters and contributes to fine-tuning. The use of PSO also enhances the robustness and generalization ability of the CNN model and its iterative nature allows for efficient discovery of hyperparameter similarity. The model was trained and tested on 13 different benchmark datasets and evaluated through six metrics: mean absolute error (MAE), mean square error (MSE), mean magnitude relative error (MMRE), root mean square error (RMSE), median magnitude relative error (MdMRE), and prediction accuracy (PRED). Comparative results reveal that the performance of the proposed model is better than other methods for all datasets and evaluation criteria. The results were very promising for predicting software cost estimation.

On-tree fruit image segmentation comparing Mask R-CNN and Vision Transformer models. Application in a novel algorithm for pixel-based fruit size estimation

In situ automatic fruit monitoring is of great interest for more accurate and cost-efficient decision making in agriculture. For this purpose, the development of computer vision-based tools is essential. Deep Learning techniques have shown good performance in fruit detection and segmentation. Recently, new models based on Transformer architecture have emerged with promising potential and zero-shot inference capability. In this paper, a Deep Learning model, Mask R-CNN, was trained for on-tree pomegranate fruit segmentation and compared with foundational models based on Vision Transformer, Grounding DINO and Segment Anything Model. Results with Mask R-CNN proved a better performance, according to F1 score and AP metrics, and a lower computational cost, according to prediction time. One of the most interesting derived applications from fruit segmentation is fruit size estimation. However, segmented fruit masks are frequently incomplete due to occlusions. Therefore, image fruit size estimation is not a straightforward process. In this work, we also propose a novel algorithm to estimate and monitor the fruit size in pixel units from the automated masks. A median relative error of 1.39% was obtained, demonstrating the potential and feasibility of future fully-automatic fruit size estimators.

Graph network surrogate model for subsurface flow optimization

The optimization of well locations and controls is an important step in the design of subsurface flow operations such as oil production or geological CO2 storage. These optimization problems can be computationally expensive, however, as many potential candidate solutions must be evaluated. In this study, we propose a graph network surrogate model (GNSM) for optimizing well placement and controls. The GNSM transforms the flow model into a computational graph that involves an encoding-processing-decoding architecture. Separate networks are constructed to provide global predictions for the pressure and saturation state variables. Model performance is enhanced through the inclusion of the single-phase steady-state pressure solution as a feature. A multistage multistep strategy is used for training. The trained GNSM is applied to predict flow responses in a 2D unstructured model of a channelized reservoir. Results are presented for a large set of test cases, in which five injection wells and five production wells are placed randomly throughout the model, with a random control variable (bottom-hole pressure) assigned to each well. Median relative error in pressure and saturation for 300 such test cases is 1-2%. The ability of the trained GNSM to provide accurate predictions for new (geologically similar) permeability realizations is demonstrated. Finally, the trained GNSM is used to optimize well locations and controls with a differential evolution algorithm. GNSM-based optimization results are comparable to those from simulation-based optimization, with a runtime speedup factor of 36. Much larger speedups are expected if the method is used for robust optimization, in which each candidate solution is evaluated on multiple geological models.

Development of a convolutional neural network based regional flood frequency analysis model for South-east Australia

Flood is one of the worst natural disasters, which causes significant damage to economy and society. Flood risk assessment helps to reduce flood damage by managing flood risk in flood affected areas. For ungauged catchments, regional flood frequency analysis (RFFA) is generally used for design flood estimation. This study develops a Convolutional Neural Network (CNN) based RFFA technique using data from 201 catchments in south-east Australia. The CNN based RFFA technique is compared with multiple linear regression (MLR), support vector machine (SVM), and decision tree (DT) based RFFA models. Based on a split-sample validation using several statistical indices such as relative error, bias and root mean squared error, it is found that the CNN model performs best for annual exceedance probabilities (AEPs) in the range of 1 in 5 to 1 in 100, with median relative error values in the range of 29–44%. The DT model shows the best performance for 1 in 2 AEP, with a median relative error of 24%. The CNN model outperforms the currently recommended RFFA technique in Australian Rainfall and Runoff (ARR) guideline. The findings of this study will assist to upgrade RFFA techniques in ARR guideline in near future.

Age Determination of LAMOST Red Giant Branch Stars Based on the Gradient Boosting Decision Tree Method

In this study, we estimate the stellar ages of LAMOST DR8 red giant branch (RGB) stars based on the gradient boosting decision tree (GBDT) algorithm. We used 2643 RGB stars extracted from the APOKASC-2 asteroseismological catalog as the training data set. After selecting the parameters ([α/Fe], [C/Fe], T eff, [N/Fe], [C/H], log g) highly correlated with age using GBDT, we apply the same GBDT method to the new catalog of more than 590,000 stars classified as RGB stars. The test data set shows that the median relative error is around 11.6% for the method. We also compare the predicted ages of RGB stars with other studies (e.g., based on APOGEE) and find some systematic differences. The final uncertainty is about 15%–30% compared to the ages of open clusters. Then, we present the spatial distribution of the RGB sample with an age determination, which could recreate the expected result, and discuss systematic biases. All these diagnostics show that one can apply the GBDT method to other stellar samples to estimate atmospheric parameters and age.

Generation and applications of synthetic computed tomography images for neurosurgical planning.

CT and MRI are synergistic in the information provided for neurosurgical planning. While obtaining both types of images lends unique data from each, doing so adds to cost and exposes patients to additional ionizing radiation after MRI has been performed. Cross-modal synthesis of high-resolution CT images from MRI sequences offers an appealing solution. The authors therefore sought to develop a deep learning conditional generative adversarial network (cGAN) which performs this synthesis. Preoperative paired CT and contrast-enhanced MR images were collected for patients with meningioma, pituitary tumor, vestibular schwannoma, and cerebrovascular disease. CT and MR images were denoised, field corrected, and coregistered. MR images were fed to a cGAN that exported a "synthetic" CT scan. The accuracy of synthetic CT images was assessed objectively using the quantitative similarity metrics as well as by clinical features such as sella and internal auditory canal (IAC) dimensions and mastoid/clinoid/sphenoid aeration. A total of 92,981 paired CT/MR images obtained in 80 patients were used for training/testing, and 10,068 paired images from 10 patients were used for external validation. Synthetic CT images reconstructed the bony skull base and convexity with relatively high accuracy. Measurements of the sella and IAC showed a median relative error between synthetic CT scans and ground truth images of 6%, with greater variability in IAC reconstruction compared with the sella. Aerations in the mastoid, clinoid, and sphenoid regions were generally captured, although there was heterogeneity in finer air cell septations. Performance varied based on pathology studied, with the highest limitation observed in evaluating meningiomas with intratumoral calcifications or calvarial invasion. The generation of high-resolution CT scans from MR images through cGAN offers promise for a wide range of applications in cranial and spinal neurosurgery, especially as an adjunct for preoperative evaluation. Optimizing cGAN performance on specific anatomical regions may increase its clinical viability.

Surrogate model for geological CO[formula omitted] storage and its use in hierarchical MCMC history matching

Deep-learning-based surrogate models show great promise for use in geological carbon storage operations. In this work we target an important application — the history matching of storage systems characterized by a high degree of (prior) geological uncertainty. Toward this goal, we extend the recently introduced recurrent R-U-Net surrogate model to treat geomodel realizations drawn from a wide range of geological scenarios. These scenarios are defined by a set of metaparameters, which include the horizontal correlation length, mean and standard deviation of log-permeability, permeability anisotropy ratio, and constants in the porosity-permeability relationship. An infinite number of realizations can be generated for each set of metaparameters, so the range of prior uncertainty is large. The surrogate model is trained with flow simulation results, generated using the open-source simulator GEOS, for 2000 random realizations. The flow problems involve four wells, each injecting 1 Mt CO2/year, for 30 years. The trained surrogate model is shown to provide accurate predictions for new realizations over the full range of geological scenarios, with median relative error of 1.3% in pressure and 4.5% in saturation. The surrogate model is incorporated into a hierarchical Markov chain Monte Carlo history matching workflow, where the goal is to generate history matched geomodel realizations and posterior estimates of the metaparameters. We show that, using observed data from monitoring wells in synthetic ‘true’ models, geological uncertainty is reduced substantially. This leads to posterior 3D pressure and saturation fields that display much closer agreement with the true-model responses than do prior predictions.

A deep-learning approach to the 3D reconstruction of dust density and temperature in star-forming regions

Aims. We introduce a new deep-learning approach for the reconstruction of 3D dust density and temperature distributions from multi-wavelength dust emission observations on the scale of individual star-forming cloud cores (&lt;0.2 pc). Methods. We constructed a training data set by processing cloud cores from the Cloud Factory simulations with the POLARIS radiative transfer code to produce synthetic dust emission observations at 23 wavelengths between 12 and 1300 µm. We simplified the task by reconstructing the cloud structure along individual lines of sight (LoSs) and trained a conditional invertible neural network (cINN) for this purpose. The cINN belongs to the group of normalising flow methods and it is able to predict full posterior distributions for the target dust properties. We tested different cINN setups, ranging from a scenario that includes all 23 wavelengths down to a more realistically limited case with observations at only seven wavelengths. We evaluated the predictive performance of these models on synthetic test data. Results. We report an excellent reconstruction performance for the 23-wavelength cINN model, achieving median absolute relative errors of about 1.8% in log(n/m−3) and 1% in log(Tdust/K), respectively. We identify trends towards an overestimation at the low end of the density range and towards an underestimation at the high end of both the density and temperature values, which may be related to a bias in the training data. After limiting our coverage to a combination of only seven wavelengths, we still find a satisfactory performance with average absolute relative errors of about 2.8% and 1.7% in log(n/m−3) and log(Tdust/K). Conclusions. This proof-of-concept study shows that the cINN-based approach for 3D reconstruction of dust density and temperature is very promising and it is even compatible with a more realistically constrained wavelength coverage.

The Measurement of Masses of OB-type Stars from LAMOST DR5

The measurements of masses and luminosities of massive stars play an important role in understanding the formation and evolution of their host galaxies. In this work, we present the measurement of masses and luminosities of 2946 OB-type stars, including 78 O-type stars and 2868 B-type stars, based on their stellar parameters (effective temperature, surface gravity, and metallicity) and the Padova and Trieste Stellar Evolution Code isochrones model. Our results show that the median mass and luminosity of the 2946 OB-type stars are 5.4 M ⊙ and log(L/L ⊙) = 3.2 with median relative errors of 21.4% and 71.1%, respectively. A good agreement between our results estimated by using our method and those derived by using the orbital motions of binary stars from the literature is found for some B-type stars. In addition, we also fit the mass–luminosity relation of B-type stars by using our derived mass and the luminosity from Gaia Data Release 3.

Peaks-Over-Threshold-Based Regional Flood Frequency Analysis Using Regularised Linear Models

Regional flood frequency analysis (RFFA) is widely used to estimate design floods in ungauged catchments. Most of the RFFA techniques are based on the annual maximum (AM) flood model; however, research has shown that the peaks-over-threshold (POT) model has greater flexibility than the AM model. There is a lack of studies on POT-based RFFA techniques. This paper presents the development of POT-based RFFA techniques, using regularised linear models (least absolute shrinkage and selection operator, ridge regression and elastic net regression). The results of these regularised linear models are compared with multiple linear regression. Data from 145 stream gauging stations of south-east Australia are used in this study. A leave-one-out cross-validation is adopted to compare these regression models. It has been found that the regularised linear models provide quite accurate flood quantile estimates, with a median relative error in the range of 37 to 47%, which outperform the AM-based RFFA techniques currently recommended in the Australian Rainfall and Runoff guideline. The developed RFFA technique can be used to estimate flood quantiles in ungauged catchments in the study region.

Accuracy of the analytical demagnetization tensor for various geometries

We investigate the accuracy of the analytical expressions for the magnetostatic demagnetization tensor for three different geometries by comparing these to the dipole field at far-away distances. We consider a prism, tetrahedron and cylindrical tile geometry. We show that for the prism and tetrahedron tiles the median relative error compared to the dipole field reaches below 10−7 at 200 tile radii away from the generating tile, while the cylindrical tile has a median relative error of about 10−5 at this distance. Even at a distance of 104 tile radii the relative error is below 10−2 for all tiles, also when single precision is used to calculate the magnetic field. This shows that the demagnetization tensor is an accurate way of calculating the stray magnetic field also for far-away distances.

Using age compositions derived from spatio-temporal models and acoustic data collected by uncrewed surface vessels to estimate Pacific hake (Merluccius productus) biomass-at-age

Generating biomass-at-age indices for fisheries stock assessments with acoustic data collected by uncrewed surface vessels (USVs) has been hampered by the need to resolve acoustic backscatter with contemporaneous biological (e.g., age) composition data. To address this limitation, Pacific hake (Merluccius productus; “hake”) acoustic data were gathered from a USV survey (in 2019) and acoustic-trawl survey (ATS; 2019 and eight previous years), and biological data were gathered from fishery-dependent and non-target (i.e., not specifically targeting hake) fishery-independent sources (2019 and eight previous years). To overcome the lack of contemporaneous biological sampling in the USV survey, age class compositions were estimated from a generalized linear mixed spatio-temporal model (STM) fit to the fishery-dependent and non-target fishery-independent data. The validity of the STM age composition estimation procedure was assessed by comparing estimates to age compositions from the ATS in each year. Hake biomass-at-age was estimated from all combinations of acoustics (USV or ATS in 2019, ATS only in other years) and age composition information (STM or ATS in all years). Across the survey area, proportional age class compositions derived from the best STM differed from ATS observations by 0.09 on average in 2019 (median relative error (MRE): 19.45%) and 0.14 across all years (MRE: 79.03%). In data-rich areas (i.e., areas with regular fishery operations), proportional age class compositions from the STM differed from ATS observations by 0.03 on average in 2019 (MRE: 11.46%) and 0.09 across years (MRE: 54.96%). On average, total biomass estimates derived using STM age compositions differed from ATS age composition-based estimates by approximately 7% across the study period (~ 3% in 2019) given the same source of acoustic data. When biomass estimates from different sources of acoustic data (USV or ATS) were compared given the same source of age composition data, differences were nearly ten-fold greater (22% or 27%, depending on if ATS or STM age compositions were used). STMs fit to non-contemporaneous data may provide suitable information for assigning population structure to acoustic backscatter in data-rich areas, but advancements in acoustic data processing (e.g., automated echo classification) may be needed to generate viable USV-based estimates of biomass-at-age.

AllCCS2: Curation of Ion Mobility Collision Cross-Section Atlas for Small Molecules Using Comprehensive Molecular Representations.

The development of ion mobility-mass spectrometry (IM-MS) has revolutionized the analysis of small molecules, such as metabolomics, lipidomics, and exposome studies. The curation of comprehensive reference collision cross-section (CCS) databases plays a pivotal role in the successful application of IM-MS for small-molecule analysis. In this study, we presented AllCCS2, an enhanced version of AllCCS, designed for the universal prediction of the ion mobility CCS values of small molecules. AllCCS2 incorporated newly available experimental CCS data, including 10,384 records and 7713 unified values, as training data. By leveraging a neural network trained on diverse molecular representations encompassing mass spectrometry features, molecular descriptors, and graph features extracted using a graph convolutional network, AllCCS2 achieved exceptional prediction accuracy. AllCCS2 achieved median relative error (MedRE) values of 0.31, 0.72, and 1.64% in the training, validation, and testing sets, respectively, surpassing existing CCS prediction tools in terms of accuracy and coverage. Furthermore, AllCCS2 exhibited excellent compatibility with different instrument platforms (DTIMS, TWIMS, and TIMS). The prediction uncertainties in AllCCS2 from the training data and the prediction model were comprehensively investigated by using representative structure similarity and model prediction variation. Notably, small molecules with high structural similarities to the training set and lower model prediction variation exhibited improved accuracy and lower relative errors. In summary, AllCCS2 serves as a valuable resource to support applications of IM-MS technologies. The AllCCS2 database and tools are freely accessible at http://allccs.zhulab.cn/.