Genetic prediction of ICU hospitalization and mortality in COVID-19 patients using artificial neural networks.

There is an unmet need of models for early prediction of morbidity and mortality of Coronavirus disease-19 (COVID-19). We aimed to a) identify complement-related genetic variants associated with the clinical outcomes of ICU hospitalization and death, b) develop an artificial neural network (ANN) predicting these outcomes and c) validate whether complement-related variants are associated with an impaired complement phenotype. We prospectively recruited consecutive adult patients of Caucasian origin, hospitalized due to COVID-19. Through targeted next-generation sequencing, we identified variants in complement factor H/CFH, CFB, CFH-related, CFD, CD55, C3, C5, CFI, CD46, thrombomodulin/THBD, and A Disintegrin and Metalloproteinase with Thrombospondin motifs (ADAMTS13). Among 381 variants in 133 patients, we identified 5 critical variants associated with severe COVID-19: rs2547438 (C3), rs2250656 (C3), rs1042580 (THBD), rs800292 (CFH) and rs414628 (CFHR1). Using age, gender and presence or absence of each variant, we developed an ANN predicting morbidity and mortality in 89.47% of the examined population. Furthermore, THBD and C3a levels were significantly increased in severe COVID-19 patients and those harbouring relevant variants. Thus, we reveal for the first time an ANN accurately predicting ICU hospitalization and death in COVID-19 patients, based on genetic variants in complement genes, age and gender. Importantly, we confirm that genetic dysregulation is associated with impaired complement phenotype.

Genetic justification of severe COVID-19 using a rigorous algorithm.

Recent studies suggest excessive complement activation in severe coronavirus disease-19 (COVID-19). The latter shares common characteristics with complement-mediated thrombotic microangiopathy (TMA). We hypothesized that genetic susceptibility would be evident in patients with severe COVID-19 (similar to TMA) and associated with disease severity. We analyzed genetic and clinical data from 97 patients hospitalized for COVID-19. Through targeted next-generation-sequencing we found an ADAMTS13 variant in 49 patients, along with two risk factor variants (C3, 21 patients; CFH,34 patients). 31 (32%) patients had a combination of these, which was independently associated with ICU hospitalization (p = 0.022). Analysis of almost infinite variant combinations showed that patients with rs1042580 in thrombomodulin and without rs800292 in complement factor H did not require ICU hospitalization. We also observed gender differences in ADAMTS13 and complement-related variants. In light of encouraging results by complement inhibitors, our study highlights a patient population that might benefit from early initiation of specific treatment.

A combination of fuzzy Delphi method and hybrid ANN-based systems to forecast ground vibration resulting from blasting.

This study presents a new input parameter selection and modeling procedure in order to control and predict peak particle velocity (PPV) values induced by mine blasting. The first part of this study was performed through the use of fuzzy Delphi method (FDM) to identify the key input variables with the deepest influence on PPV based on the experts' opinions. Then, in the second part, the most effective parameters on PPV were selected to be applied in hybrid artificial neural network (ANN)-based models i.e., genetic algorithm (GA)-ANN, particle swarm optimization (PSO)-ANN, imperialism competitive algorithm (ICA)-ANN, artificial bee colony (ABC)-ANN and firefly algorithm (FA)-ANN for the prediction of PPV. Many hybrid ANN-based models were constructed according to the most influential parameters of GA, PSO, ICA, ABC and FA optimization techniques and 5 hybrid ANN-based models were proposed to predict PPVs induced by blasting. Through simple ranking technique, the best hybrid model was selected. The obtained results revealed that the FA-ANN model is able to offer higher accuracy level for PPV prediction compared to other implemented hybrid models. Coefficient of determination (R2) results of (0.8831, 0.8995, 0.9043, 0.9095 and 0.9133) and (0.8657, 0.8749, 0.8850, 0.9094 and 0.9097) were obtained for train and test stages of GA-ANN, PSO-ANN, ICA-ANN, ABC-ANN and FA-ANN, respectively. The results showed that all hybrid models can be used to solve PPV problem, however, when the highest prediction performance is needed, the hybrid FA-ANN model would be the best choice.

A Novel Feature Selection Approach Based on Tree Models for Evaluating the Punching Shear Capacity of Steel Fiber-Reinforced Concrete Flat Slabs.

When designing flat slabs made of steel fiber-reinforced concrete (SFRC), it is very important to predict their punching shear capacity accurately. The use of machine learning seems to be a great way to improve the accuracy of empirical equations currently used in this field. Accordingly, this study utilized tree predictive models (i.e., random forest (RF), random tree (RT), and classification and regression trees (CART)) as well as a novel feature selection (FS) technique to introduce a new model capable of estimating the punching shear capacity of the SFRC flat slabs. Furthermore, to automatically create the structure of the predictive models, the current study employed a sequential algorithm of the FS model. In order to perform the training stage for the proposed models, a dataset consisting of 140 samples with six influential components (i.e., the depth of the slab, the effective depth of the slab, the length of the column, the compressive strength of the concrete, the reinforcement ratio, and the fiber volume) were collected from the relevant literature. Afterward, the sequential FS models were trained and verified using the above-mentioned database. To evaluate the accuracy of the proposed models for both testing and training datasets, various statistical indices, including the coefficient of determination (R2) and root mean square error (RMSE), were utilized. The results obtained from the experiments indicated that the FS-RT model outperformed FS-RF and FS-CART models in terms of prediction accuracy. The range of R2 and RMSE values were obtained as 0.9476-0.9831 and 14.4965-24.9310, respectively; in this regard, the FS-RT hybrid technique demonstrated the best performance. It was concluded that the three hybrid techniques proposed in this paper, i.e., FS-RT, FS-RF, and FS-CART, could be applied to predicting SFRC flat slabs.