1
Massively-multitask regression models (MMRMs) have revolutionized activity prediction for drug discovery. MMRMs trained on millions of compounds and many thousands of assays can predict bioactivity with accuracy comparable to 4-concentration IC50 experiments. This report compares six MMRMs: pQSAR, Alchemite, MT-DNN, MetaNN, Macau and IMC. Models were trained by experts in each method, on identical sets of 159 kinase and 4276 diverse ChEMBL assays, employing realistically novel training/test set splits. Results were compared both qualitatively and with statistical rigor. Our use-case is imputing full bioactivity profiles for the very sparse compound collections on which the models were trained. MMRMs performed much better than the single-task random forest regression (ST-RFR) model. Five MMRMs train all models simultaneously, so must leave out test-set measurements from all assays to avoid leakage (here 25% of data), whereas one method trains models one-at-a-time, so only holds out test data for that assay (< 1% of data). Thus, all algorithms were compared both using 75/25 splits, and when possible, 99 + / < 1 splits. Many MMRM evaluations achieved similar accuracy when tested on the same split. However, when evaluated on 75/25 splits, all MMRMs performed much worse than when evaluated on 99 + / < 1% splits. Thus, while many MMRMs produce comparable final production models (trained on all the data), models that require 75/25 splits greatly underestimate the accuracy of the final models. While outstanding for imputations, MMRMs proved little better than ST-RFR for compounds very unlike the training collection. Thus, MMRMs are best for hit-finding, off-target, promiscuity, MoA, polypharmacology or drug-repurposing within the training collection. Since accuracy is not a deciding factor, other pros and cons of each method are also described.
Editor's evaluation: Derivation and external validation of clinical prediction rules identifying children at risk of linear growth faltering
Clinical prediction rules could help identify children at risk of slowed growth after an episode of diarrheal illness, and these rules may be generalizable to all children, regardless of diarrhea status.
Read moreExamining the utility of digital phenotyping for the prediction of intrusive experiences
Background: Intrusive experiences related to witnessing a traumatic event are the core symptom of post-traumatic stress disorder (PTSD), and have been shown to be predicted by peritraumatic emotional arousal. However, research into the role of peritraumatic arousal in the development of intrusions has been limited by a reliance on self-report scales. Objective: This study aimed to examine whether facial, focal, and language phenotypes of peritraumatic emotional arousal could also contribute to the prediction of intrusive experiences. Method: This longitudinal study recruited university students (N = 81), who were recorded describing their response to an analogue trauma (car accident video). Participants’ facial, vocal, and linguistic features were extracted from these recordings, as well as their self-reported emotional experiences, and input into machine learning (random forest regression) models to predict whether they would have intrusive experiences of the analogue trauma three days after their exposure to it. Results: Random forest regression models were able to predict variance in responsiveness to triggers that reminded individuals of the analogue trauma (R 2 = 0.23) and whether they had vivid memories of the analogue trauma (R 2 = 0.55). Vocal and language features made the greatest contribution to the prediction of responsiveness to triggers, while initial self-reported memory vividness made the greatest contribution to the prediction of three-day memory vividness. Conclusions: These findings suggest that facial, vocal, and language phenotypes, in combination with self-report measures, can have utility for predicting the occurrence of specific intrusive experiences following exposure to an analogue trauma in a university sample. Further research into this approach in clinical samples is required to demonstrate its utility for predicting intrusions in those with PTSD.
Read more3A Stable Passenger Flow Forecast Approach for Newly Opened Metro Stations Based on Multi-Source Data and Random Forest Regression Model
Forecasting the stable passenger flow of a metro station before its official operation can help the management department design a more efficient passenger flow organization scheme. However, the conventional methods based on four-step method of traffic planning have disadvantages such as a long forecasting period, low forecasting accuracy, and difficulty in obtaining data. On the other hand, because emerging big data forecasting methods are all driven by historical data, it is impossible to apply these methods before the official operation of stations. Therefore, this paper proposes a passenger flow forecast approach through multi-source data and random forest regression (RFR) model. These multi-source data mainly include the number of POI (Point of Interest), employment positions, and residents in the radiation area of the station, as well as the attributes and functions of the station. This paper obtained the multi-source data of 65 stations in Qingdao Metro for experimental evaluation. In the case study, this paper uses five models including random forest regression, linear regression, Bayesian ridge regression, elastic network, and gradient boosting regression for comparison. The results show that RFR has the most satisfactory accuracy. This study provides a new perspective for relevant management departments.
Read more180Estimation of heavy metals using deep neural network with visible and infrared spectroscopy of soil
Estimation of heavy metals using deep neural network with visible and infrared spectroscopy of soil
34Interpreting Highly Variable Indoor PM2.5 in Rural North China Using Machine Learning.
Household air pollution associated with solid fuel use is a long-standing public concern. The global population mainly using solid fuels for cooking remains large. Besides cooking, large amounts of coal and biomass fuels are burned for space heating during cold seasons in many regions. In this study, a wintertime multiple-region field campaign was carried out in north China to evaluate indoor PM2.5 variations. With hourly resolved data from ∼1600 households, key influencing factors of indoor PM2.5 were identified from a machine learning approach, and a random forest regression (RFR) model was further developed to quantitatively assess the impacts of household energy transition on indoor PM2.5. The indoor PM2.5 concentration averaged at 120 μg/m3 but ranged from 16 to ∼400 μg/m3. Indoor PM2.5 was ∼60% lower in families using clean heating approaches compared to those burning traditional coal or biomass fuels. The RFR model had a good performance (R2 = 0.85), and the interpretation was consistent with the field observation. A transition to clean coals or biomass pellets can reduce indoor PM2.5 by 20%, and further switching to clean modern energies would reduce it an additional 30%, suggesting many significant benefits in promoting clean transitions in household heating activities.
Read moreMechanistic insights into sulfamethoxazole removal in activated sludge systems through machine learning and microcosm experiments.
Mechanistic insights into sulfamethoxazole removal in activated sludge systems through machine learning and microcosm experiments.
Read more
19Evaluating how lodging affects maize yield estimation based on UAV observations.
Timely and accurate pre-harvest estimates of maize yield are vital for agricultural management. Although many remote sensing approaches have been developed to estimate maize yields, few have been tested under lodging conditions. Thus, the feasibility of existing approaches under lodging conditions and the influence of lodging on maize yield estimates both remain unclear. To address this situation, this study develops a lodging index to quantify the degree of lodging. The index is based on RGB and multispectral images obtained from a low-altitude unmanned aerial vehicle and proves to be an important predictor variable in a random forest regression (RFR) model for accurately estimating maize yield after lodging. The results show that (1) the lodging index accurately describes the degree of lodging of each maize plot, (2) the yield-estimation model that incorporates the lodging index provides slightly more accurate yield estimates than without the lodging index at three important growth stages of maize (tasseling, milking, denting), and (3) the RFR model with lodging index applied at the denting (R5) stage yields the best performance of the three growth stages, with R2 = 0.859, a root mean square error (RMSE) of 1086.412 kg/ha, and a relative RMSE of 13.1%. This study thus provides valuable insight into the precise estimation of crop yield and demonstra\tes that incorporating a lodging stress-related variable into the model leads to accurate and robust estimates of crop grain yield.
Read more16Comparison of random regression and repeatability models to predict breeding values from test-day records of Norwegian goats
Comparison of random regression and repeatability models to predict breeding values from test-day records of Norwegian goats
Read more54Estimating monthly average temperature by remote sensing in China
Estimating monthly average temperature by remote sensing in China
27Using machine learning to derive cloud condensation nuclei number concentrations from commonly available measurements
Abstract. Cloud condensation nuclei (CCN) number concentrations are an important aspect of aerosol–cloud interactions and the subsequent climate effects; however, their measurements are very limited. We use a machine learning tool, random decision forests, to develop a random forest regression model (RFRM) to derive CCN at 0.4 % supersaturation ([CCN0.4]) from commonly available measurements. The RFRM is trained on the long-term simulations in a global size-resolved particle microphysics model. Using atmospheric state and composition variables as predictors, through associations of their variabilities, the RFRM is able to learn the underlying dependence of [CCN0.4] on these predictors, which are as follows: eight fractions of PM2.5 (NH4, SO4, NO3, secondary organic aerosol (SOA), black carbon (BC), primary organic carbon (POC), dust, and salt), seven gaseous species (NOx, NH3, O3, SO2, OH, isoprene, and monoterpene), and four meteorological variables (temperature (T), relative humidity (RH), precipitation, and solar radiation). The RFRM is highly robust: it has a median mean fractional bias (MFB) of 4.4 % with ≈96.33 % of the derived [CCN0.4] within a good agreement range of -60%<MFB<+60% and strong correlation of Kendall's τ coefficient ≈0.88. The RFRM demonstrates its robustness over 4 orders of magnitude of [CCN0.4] over varying spatial (such as continental to oceanic, clean to polluted, and near-surface to upper troposphere) and temporal (from the hourly to the decadal) scales. At the Atmospheric Radiation Measurement Southern Great Plains observatory (ARM SGP) in Lamont, Oklahoma, United States, long-term measurements for PM2.5 speciation (NH4, SO4, NO3, and organic carbon (OC)), NOx, O3, SO2, T, and RH, as well as [CCN0.4] are available. We modify, optimize, and retrain the developed RFRM to make predictions from 19 to 9 of these available predictors. This retrained RFRM (RFRM-ShortVars) shows a reduction in performance due to the unavailability and sparsity of measurements (predictors); it captures the [CCN0.4] variability and magnitude at SGP with ≈67.02 % of the derived values in the good agreement range. This work shows the potential of using the more commonly available measurements of PM2.5 speciation to alleviate the sparsity of CCN number concentrations' measurements.
Read more33Analysis of Financing Efficiency of Chinese Agricultural Listed Companies Based on Machine Learning
Agricultural enterprises play a significant role in China’s economic development. However, compared with other enterprises, agricultural enterprises are facing serious financial problems. Financing difficulty is essentially a question of financing efficiency. Based on the DEA method, this paper evaluates the financing efficiency of 39 agricultural listed companies in China from 2013 to 2017. The results suggest that the financing efficiency is generally low, and the Total Factor Productivity of agricultural enterprises’ financing has a tendency to decrease first and then increase. The influencing factors of financing efficiency are analyzed using the Tobit regression model and the random forest regression model. And we find the following: (1) The random forest regression model significantly outperformed the Tobit regression model, with determination coefficients (R2) greater than 0.9 in full sample sets. (2) Total liability, financial expenses, return on total assets, and inventory turnover rate are important factors affecting financing efficiency of agricultural listed companies. (3) Return on total assets and inventory turnover rate promote the financing efficiency, while total liability and financial expenses reduce financing efficiency. Finally, the paper makes some suggestions for the financing of agricultural enterprises.
Read more8Synergy of optical and synthetic aperture radar data for early-stage crop yield estimation: a case study over a state of Germany
Traditional crop cutting experiment-based yield estimation method captures the regional yield variability but lacks field-level information. Satellite images hold enormous crop information at finer spatial resolution. Crop yield mapping with optical images is particularly challenging if cloud-free images are unavailable during the crucial crop developmental stages. All-weather availability and sensitivity to crop structure, dielectric properties make synthetic aperture radar (SAR) images an excellent resource for yield estimation. Both types of data provide complementary information about crop conditions. A random forest regression model with genetic algorithm-based feature selection is developed to exploit the Sentinel-2 optical and Sentinel-1 SAR images for yield estimation. We utilized the crop harvest and quality survey (BEE) yield data set collected by the Hessisches Statistisches Landesamt (HSL), Wiesbaden, Germany, over 490 fields. We prepared 20 m resolution yield maps for winter wheat, winter barley, winter rye and winter rapeseed. Input features for the yield estimation model are selected based on the prior knowledge of remote sensing of vegetation. Baseline random forest regression models are developed for all the four crop types with optical and SAR input features. An optimized random forest regression model with genetic algorithm-based feature selection results in performance improvement. Dissimilarity in genetic algorithm selected image features highlights the significance of crop-specific feature selection for yield estimation. The optimized models reliably estimate yield by achieving correlation coefficient (r) of 0.65–0.86, mean absolute error 0.93–1.16 t ha–1 and root mean square error 1.12–1.56 t ha–1 with BEE yield on testing data set. The proposed models could estimate the intra-field yield variation when winter wheat, winter barley, winter rye were in the shooting phase to the beginning of ear-shifting, and winter rapeseed began to flower or was already flowering. These results demonstrate the merits of our model for early-stage crop yield estimation at the field level with mono-temporal image and adaptability for the cropping season with high cloud cover.
Read more3Characterization of naturally occurring radioactive material dynamics in community water systems using groundwater from Ganghwa Island, Republic of Korea
Characterization of naturally occurring radioactive material dynamics in community water systems using groundwater from Ganghwa Island, Republic of Korea
Read moreUse of statistical models for predicting oral health status of children with cerebral palsy in Sri Lanka
Cerebral Palsy (CP) is the most common movement disorder in children, which is defined as ‘‘a group of permanent disorders of the development of movement and posture, causing activity limitations attributed to non-progressive disturbances occurred in developing fetal or infant brain. In this study, we consider the four most common CP types categorized by the location of movement problems named Monoplegia, Diplegia, Hemiplegia, and Quadriplegia. Oral health is a state of being free from the chronic mouth, facial pain, oral and throat cancer, oral sores, congenital disabilities such as cleft lip and palate, tooth decay and tooth loss, and other diseases disorders oral cavity. The main goal of the study is to create suitable statistical models for predicting the oral health status of children with CP using Silness-Löe plaque index and DMFT Index (DMFTI). Also, to identify the relationships between DMFTI and demographic, DMFTI and CP location, Silness-Loe plaque index and demographic data, Silness-Loe plaque index and CP location, Care index (CI) and demographic data, and the CI and CP location. This analysis was performed on a sample of 93 children with CP in the Central Province, Sri Lanka. The independent sample t-test and one-way ANOVA test were used to identify the relationship between variables, and effect sizes were calculated using partial Eta squared value to measure the strength of the relationship. Further Multiple Linear Regression (MLR) model, Random Forest Regression (RFR) model, and the Support Vector Regression (SVR) model were used to predict the oral health status using DMFTI and plaque index separately. A comparison was conducted for the fitted models using the Coefficient of determination (R-squared). There is a significant difference between the mean values of the plaque index for different CP locations. Children with diplegia have the lowest plaque index, while children with hemiplegia have the highest plaque index. The accuracy of the MLR model for predicting DMFTI is 23.60% and 20.80% for Permanent and primary teeth separately, and 20.00% for predicting Plaque Index. Those accuracies for the RFR model are 92.64%, 93.11% and 90.32%, while 95.36%, 85.65% and 80.07% for SVR model respectively. Therefore, the RFR Model was considered the best-fitted model for predicting oral health status using DMFTI and the plaque index of Sri Lankan children with CP. Besides, children with hemiplegia have a higher risk of having lower oral health status.
Read more50Towards safer streets: A framework for unveiling pedestrians’ perceived road safety using street view imagery
Towards safer streets: A framework for unveiling pedestrians’ perceived road safety using street view imagery
Copyright 2026 Cactus Communications. All rights reserved.
