Search PubMed⌕ Search

SEARCH · Search PubMed

Results for “Predictive”

Search indexed PubMed citations on genomics, clinical trials, systematic reviews and public health. Explore titles, authors and supplied subject terms, then open the PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,189 records · Page 66Linked to original sources
PubMed2003

Predicting reliable regions in protein alignments from sequence profiles.

For applications such as comparative modelling one major issue is the reliability of sequence alignments. Reliable regions in alignments can be predicted using sub-optimal alignments of the same pair of sequences. Here we show that reliable regions in alignments can also be predicted from multiple sequence profile information alone. Alignments were created for a set of remotely related pairs of proteins using five different test methods. Structural alignments were used to assess the quality of the alignments and the aligned positions were scored using information from the observed frequencies of amino acid residues in sequence profiles pre-generated for each template structure. High-scoring regions of these profile-derived alignment scores were a good predictor of reliably aligned regions. These profile-derived alignment scores are easy to obtain and are applicable to any alignment method. They can be used to detect those regions of alignments that are reliably aligned and to help predict the quality of an alignment. For those residues within secondary structure elements, the regions predicted as reliably aligned agreed with the structural alignments for between 92% and 97.4% of the residues. In loop regions just under 92% of the residues predicted to be reliable agreed with the structural alignments. The percentage of residues predicted as reliable ranged from 32.1% for helix residues to 52.8% for strand residues. This information could also be used to help predict conserved binding sites from sequence alignments. Residues in the template that were identified as binding sites, that aligned to an identical amino acid residue and where the sequence alignment agreed with the structural alignment were in highly conserved, high scoring regions over 80% of the time. This suggests that many binding sites that are present in both target and template sequences are in sequence-conserved regions and that there is the possibility of translating reliability to binding site prediction.

Algorithms↗
PubMed2006

External validation of outcome prediction model for ureteral/renal calculi.

PURPOSE: We externally validated a previously designed neural network model to predict outcome and duration of passage for ureteral/renal calculi. The model was also evaluated using a 6 mm largest stone dimension cutoff in predicting stone outcome. MATERIALS AND METHODS: The model was previously designed on 301 patients at Albany Medical Center (free shareware from www.uroengineering.com). The model had a prediction accuracy of 86% for passage outcome and 87% for passage duration. In this study we tested the model on a separate 384 patients from 6 different external institutions to assess the prediction accuracy. All patients had a single renal/ureteral calculus by evaluation in an emergency room setting or by primary physicians and were then referred for further treatment. Model accuracy was also compared to using a 6 mm largest stone dimension cutoff in predicting the need for intervention. RESULTS: Testing on the 384 patients from all 6 external institutions revealed an outcome prediction accuracy of 88%. The area under the ROC curve was 0.9. Using a 6 mm stone size cutoff provided 79% (ROC 0.8) accuracy. The model duration of passage prediction accuracy was 80% (133 patients passed the stone, area under ROC of 0.8). CONCLUSIONS: The model provided high stone outcome prediction accuracy (ROC of 0.9 and 0.8) at the 6 external institutions, comparable to that of the design institution. The model provided higher accuracy than using only the largest stone dimension as a cutoff. Increasing experience will further assess the model's accuracy.

Adolescent↗
PubMed2001

Sonographic prediction of twin birth weight discordance.

OBJECTIVE: To assess the accuracy of sonographic prediction of clinically significant twin birth weight discordance (25% or greater) and to determine whether this accuracy is affected by defined fetal and maternal variables. METHODS: Using an established database, we reviewed 338 twin gestations delivered over 10 years as a retrospective cohort. Estimation of fetal weight was calculated by applying the Hadlock formula using composite fetal biometry. Intertwin weight discordance was calculated as the difference in the estimated or actual twin weights (A-B) divided by the weight of the larger twin and was expressed as a percentage. Statistical evaluation included validity (sensitivity, specificity, and predictive values) and reliability assessment of ultrasonographic measurements (intraclass correlation coefficients). Multivariable analysis was performed. RESULTS: Of 338 twin gestations, 192 (57%) twin pairs met inclusion criteria. Sonographic prediction of actual intertwin birth weight discordance of 25% or greater had a sensitivity of 55%, specificity of 97%, positive predictive value of 82%, and negative predictive value of 91%. The reliability of estimating intertwin birth weight discordance by ultrasonography was moderately high (intraclass correlation coefficient =.700; 95% confidence interval [CI].620,.765). Multivariable analysis revealed no significant effects of individual maternal or fetal factors on the accuracy of ultrasonographic prediction of intertwin birth weight discordance. CONCLUSION: Sonographic prediction of actual intertwin birth weight discordance of 25% or greater within 16 days of delivery appears to be a valid and reliable method for clinical use. Predictive accuracy is independent of other identifiable maternal or fetal variables.

Adolescent↗
PubMedDate not supplied

Predictability of the one-year course of depression and generalized anxiety in primary care.

Several predictors of the course of depression and generalized anxiety have been identified. Whether these predictors provide a solid basis for primary care physicians (PCPs) to give an accurate prognosis remains unclear. A parallel study showed modest agreement between PCP prognosis and observed course (kappa< or = 0.21). It is the aim of the present study to establish the extent to which the one-year course of depression and generalized anxiety in primary care is in fact predictable. Predictability is operationalized as the combined predictive power of major prognostic factors identified in the literature. We identified 269 cases of ICD-10 depression and 134 of generalized anxiety among consecutive PCP attenders. For these patients a statistical model was built that provided optimal predictions of the one-year course of the disorder, based on the prognostic factors discerned. The predictions were compared with the actual course observed. Reasonable agreement (kappa = 0.37 for depression, kappa = 0.35 for anxiety) and good association (gamma = 0.66 for depression, gamma=0.67 for anxiety) were found between predicted and observed course. Nevertheless, the combined predictive power of the prognostic factors remains limited. A realistic evaluation of the accuracy of the PCP prognosis should take this limited predictability into account.

Adult↗
PubMed2000

Evaluation of three simple methods for predicting therapeutic lithium doses.

The bias and accuracy of three simple methods for predicting lithium doses were assessed in this prospective study. In each patient, we computed the predicted doses (PDs) of lithium by applying three formulas. The actual dose (AD), the lowest lithium dose that was enough to produce a serum lithium concentration higher than 0.80 mmol/l, was determined. The PD computed by each formula was then compared with the AD to identify the one with the least bias and the greatest accuracy in predicting therapeutic lithium doses. As mean prediction error (ME) is a convenient measure of bias, the prediction error (PE) of each comparison was computed by subtracting the AD from the PD. Accuracy was assessed as a function of the root-mean-squared prediction error (rMSE). Seventeen psychiatric inpatients participated in this study. The 95% confidence interval of ME of a proposed formula [Zetin, M., Garber, D., De Antonio, M., Schlegel, A., Feureisen, S., Fieve, R., Jewett, C., Reus, V., Huey, L.Y., 1986. Prediction of lithium dose: a mathematic alternative to the test-dose method. Journal of Clinical Psychiatry 47, 175-178] was across zero (-15.48 to 133.72). In addition, the rMSE of that formula was also the lowest one (152.66 mg/day). The method proposed by Zetin et al. (1986) is the least biased and the most accurate way to predict therapeutic lithium doses.

Adolescent↗
PubMed2000

Predictive modelling of the growth and survival of Listeria in fishery products.

Predictive microbiology provides a powerful tool to aid the exposure assessment phase of 'quantitative microbial risk assessment'. Using predictive models changes in microbial populations on foods between the point of production/harvest and the point of eating can be estimated from changes in product parameters (temperature, storage atmosphere, pH, salt/water activity, etc.). Thus, it is possible to infer exposure to Listeria monocytogenes at the time of consumption from the initial microbiological condition of the food and its history from production to consumption. Predictive microbiology models have immediate practical application to improve microbial food safety and quality, and are leading to development of a quantitative understanding of the microbial ecology of foods. While models are very useful decision-support tools it must be remembered that models are, at best, only a simplified representation of reality. As such, application of model predictions should be tempered by previous experience, and used with cognisance of other microbial ecology principles that may not be included in the model. Nonetheless, it is concluded that predictive models, successfully validated in agreement with defined performance criteria, will be an essential element of exposure assessment within formal quantitative risk assessment. Sources of data and models relevant to assessment of the human health risk of L. monocytogenes in seafoods are identified. Limitations of the current generation of predictive microbiology models are also discussed. These limitations, and their consequences, must be recognised and overtly considered so that the risk assessment process remains transparent. Furthermore, there is a need to characterise and incorporate into models the extent of variability in microbial responses. The integration of models for microbial growth, growth limits or inactivation into models that can predict both increases and decreases in microbial populations over time will also improve the utility of predictive models for exposure assessment. All of these issues are the subject of ongoing research.

Animals↗
PubMed1995

Predicting outcome in pediatric submersion victims.

STUDY OBJECTIVE: To predict outcome in children after near-drowning. DESIGN: Retrospective cohort study. Vegetative state and death were classified as unfavorable outcomes, whereas all other outcomes were classified as favorable. Demographic, episode-related, clinical, laboratory, and treatment variables available at the time of admission were evaluated for their usefulness in predicting outcome. SETTING: Pediatric referral hospital. PARTICIPANTS: Children admitted after submersion injury in non-icy waters. RESULTS: The study cohort comprised 194 children (median age, 2.6 years; range, 5 months to 18 years); 131 were neurologically normal at the time of discharge, 10 had some degree of neurologic impairment, 15 were in a vegetative state at the time of discharge, and 38 died. We used a combination of partitioning and logistic regression to combine variables in a prediction rule that was always correct when unfavorable outcome was predicted. The final rule predicted favorable outcome for all children who were not comatose. Among comatose children, unfavorable outcome was predicted by a combination of absent pupillary light reflex, increased initial blood glucose concentration, and male sex. This rule had a specificity of 100%--children with favorable outcomes were always predicted to do well--and a sensitivity of 65%. Therefore the rule was overly optimistic for 35% of patients with unfavorable outcomes. CONCLUSION: Pediatric submersion victims can be assigned to high or low likelihoods of unfavorable outcome with the use of four variables: comatose state, lack of pupillary light reflex, sex, and initial blood glucose concentration. This prediction rule may be useful if it can be validated in another cohort.

Adolescent↗
PubMed1999

Can the probability for obliteration after radiosurgery for arteriovenous malformations be accurately predicted?

PURPOSE: To investigate how accurate different models predict the probability for obliteration following radiosurgery for an arteriovenous malformation (AVM). METHODS AND MATERIALS: The probability for obliteration was calculated for all 838 AVMs with a known treatment outcome and treated at the Karolinska Hospital with Gamma Knife surgery 1970-1993. Four different models were used for the calculation, resulting in four different values of the probability for obliteration. The calculated prediction values were added for each model, and the total number of predicted obliteration compared to that observed in the whole patient material as well as in different subgroups. RESULTS: Three of the four models predicted the total number of obliterations accurately. In two of those three models, the accuracy of the prediction was dependent on AVM volume and treatment dose. In one model only, the prediction was accurate and independent of all investigated parameters. CONCLUSIONS: The probability for obliteration was accurately predicted by one of the models analyzed. In this model, the probability for obliteration was related to the dose to the AVM periphery only. The AVM volume had no independent impact on the probability for obliteration. There was a trend that AVMs with a central location had a better obliteration rate than predicted.

Adult↗
PubMed2002

Prediction of axial length on the basis of vitreous body length and lens thickness: retrospective echobiometric study.

PURPOSE: To examine the predictability of axial length measurement when lens thickness and vitreous body length are known. SETTING: Center of Ophthalmology, University of Cologne, Cologne, Germany. METHODS: This study comprised 227 patients with a mean age of 70.01 years +/- 12.47 (SD). None of the eyes examined had a history of surgery or trauma. Patients with systemic disease were excluded from the study. Before cataract surgery, the length of the vitreous body and thickness of the lens were measured and the correlation between these data and the axial length was evaluated. Two ultrasonic devices were used for biometric measurements: the BMS 811 Biometric System (Grieshaber) and the Cooper Vision Ultrascan Digital A+B-Scan 2000. The correlations between the axial length and vitreous body length and the lens thickness and vitreous body length were analyzed using multiple linear regression. RESULTS: The BMS 811 provided the best prediction of axial length based on vitreous body length. Considering sex but not age significantly improved the model fit. With the BMS 811, the following formula was developed to predict axial length: Axial length (mm) = 7.129 mm + 0.095 mm x sex (female = 0, male = 1) + 1.040 x vitreous body length (mm). An approximate 95% prediction limit may be calculated by the following formula: Axial length (mm) +/- 2 x 0.413 mm. CONCLUSIONS: This study yielded an easy-to-use formula for predicting the axial length using the vitreous body length and the patient's sex. The remaining error in prediction is likely to be the result of patient heterogeneity in age, ocular globe size, and lens thickness (cataract formation). Good prediction of the axial length is important to refractive outcomes to distinguish corneal myopia from axial length myopia to choose grafts and the opening size in penetrating keratoplasty. Further studies to detect a clinically relevant improvement in such outcomes are required to assess the utility of the prediction formula.

Aged↗
PubMed2003

Predicting mandibular growth potential with cervical vertebral bone age.

This study assessed the possibility of using cervical vertebral bone age determined from cephalometric radiographs to predict mandibular growth potential. The subjects were 2 groups of 20 Japanese girls and young women: one group to derive a formula for predicting mandibular growth potential, the other to compare predicted values with actual values. Each group included subjects in the initial stage of the pubertal growth period and the final stage of growth in early adulthood. A formula for predicting mandibular growth potential that included cervical vertebral bone age and the actual growth of the mandible (condylion-gnathion) was determined with regression analysis. Cervical vertebral bone age, bone age on hand-wrist radiographs, and chronological age were inserted into the formula, and actual values and values predicted with these parameters of the formula for mandibular growth potential were compared. The formula found mandibular growth potential (in millimeters) = -2.76 x cervical vertebral bone age + 38.68. The average error between the value predicted by cervical vertebral bone age and the actual value (1.79 mm) was significantly less (P <.001) than that between the actual value and the value predicted by chronological age (3.48 mm) and approximately the same as that between the actual value and the value predicted by bone age (2.09 mm). The formula derived from this study might be useful for treating orthodontic patients in the growth stage.

Adolescent↗
PubMed1998

Multivariate prediction of skeletal Class II growth.

Prediction of craniofacial growth is one of the keys to successful orthodontic treatment and stability. Despite numerous attempts at growth forecasting, our ability to accurately predict growth is limited. The present study outlines a possible new approach to prediction of craniofacial growth that differs from any previous attempt because of both the methods used and type of patients studied. The purpose of this study is to create and test prediction equations for forecasting favorable or unfavorable patterns of growth in skeletal Class II preadolescents. The subjects for this study include 19 females and 12 males from the Bolton growth center in Cleveland, Ohio. The patients were all untreated orthodontically, had lateral cephalometric headfilms taken biannually from the ages of 6 through 18 and had a Class II skeletal relationship at age 8. Twenty-six skeletal and dental landmarks were identified and digitized, and 48 linear, angular, and proportional measurements were calculated. The subjects were divided into two groups based on anterior-posterior changes in the relationship between the maxilla and mandible. Eleven patients were in the favorable growth group, with an average improvement of 4.13 degrees in the ANB angle; 20 patients were in the unfavorable growth group with an average increase of 0.16 degrees in the ANB angle. The following prediction formula was created with Bayes theorem and assuming a multivariate Gaussian distribution: P(Good¿Fn) = ke (-(0.5)) ¿Fn - mu(ng)¿sigma(g)(-1)¿Fn - mu(ng)¿T. The equation's sensitivity and specificity was calculated from serial cephalometric data from ages 6, 8, 10, and 12. The results obtained with this equation indicate 82.2% sensitivity, 95% specificity with a overall positive predictive value of 91%. This corresponds to 17.8% of patients being incorrectly identified as Poor Growers and only 5% of our patients were incorrectly identified as Good Growers. We conclude that this prediction formula improves the ability to predict favorable or unfavorable patterns of growth in this sample of skeletal Class II preadolescents.

Adolescent↗
PubMed1999

Evaluation of Ricketts' long-range growth prediction in Turkish children.

This study was conducted to evaluate Ricketts' long-range growth prediction in Turkish children. Cephalometric analysis was conducted at baseline and 7 years for 40 children (20 girls, 20 boys) who received no orthodontic treatment. Ricketts' long-range prediction was performed from baseline cephalograms and compared with actual growth 7 years later. Twenty-one cephalometric (12 angular and 9 linear) parameters were measured on actual and predicted tracings. The Pearson correlation coefficient was used to evaluate relationships between the "predicted" and "actual" measurements. Analysis was conducted on pooled data (males and females) and data by sex. There was a higher level of correlation for growth prediction in girls. Data indicate predictability in boys was greater for maxillary mandibular growth parameters. It was concluded that Ricketts' long-range growth prediction may be helpful in improving treatment planning. Further work on accurate soft tissue and hard tissue growth prediction is indicated.

Cephalometry↗
PubMed2003

Prediction precedes control in motor learning.

Skilled motor behavior relies on the brain learning both to control the body and predict the consequences of this control. Prediction turns motor commands into expected sensory consequences, whereas control turns desired consequences into motor commands. To capture this symmetry, the neural processes underlying prediction and control are termed the forward and inverse internal models, respectively. Here, we investigate how these two fundamental processes are related during motor learning. We used an object manipulation task in which subjects learned to move a hand-held object with novel dynamic properties along a prescribed path. We independently and simultaneously measured subjects' ability to control their actions and to predict their consequences. We found different time courses for predictor and controller learning, with prediction being learned far more rapidly than control. In early stages of manipulating the object, subjects could predict the consequences of their actions, as measured by the grip force they used to grasp the object, but could not generate appropriate actions for control, as measured by their hand trajectory. As predicted by several recent theoretical models of sensorimotor control, our results indicate that people can learn to predict the consequences of their actions before they can learn to control their actions.

Biophysical Phenomena↗
PubMed1998

Accurate prediction of term birth weight from prospectively measurable maternal characteristics.

Objective: To determine whether accurate prediction of term birth weight is possible based on maternal characteristics routinely measured remote from term, and to compare this technique to more expensive methods requiring obstetrical ultrasound examinations.Methods: Two hundred fifty-nine normal, non-smoking, non-diabetic Caucasian gravidas with uncomplicated term gestations were studied. Seven maternal characteristics (age, parity, height, weight, level of obesity, rate of pregnancy weight gain, and glucose screening test result) and two fetal characteristics (fetal gender and length of gestation) were evaluated alone and in combination for their predictive values in determining birth weight. A cross-validated split-sample multiple regression analysis was used to determine which combinations of these variables were significant and a birth weight prediction equation was developed. Predictive accuracy was assessed using a jackknifing procedure, and results were compared to similar types of birthweight predictions obtained both via previous algorithms based upon maternal characteristics and those developed for use with fetal ultrasonographic biometric data.Results: Significant predictors of term birth weight were gestational age, fetal gender, parity, maternal height, maternal weight, and third trimester maternal weight gain rate. These prospectively measurable variables explained 33% of the variance in birth weight and predicted birth weight to within 10.8%. These results were compared to those obtained from other previously published algorithms and were more accurate than all others derived from either maternal characteristics or fetal ultrasonographic data. Our term birth weight prediction equation is: birth weight (g)=gestational age (days)x[9.40+0.255xgender+ 0.000232xheight (cm)xmaternal weight at 26 wk (kg)+ 4.89x3rd trimester weight gain rate (kg/d)x(parity+1)]where: gender=-1 for females;+1 for males;0 for unknown gender gestational age=conceptual age (days)+14Conclusion: Term birth weight can be accurately predicted using routinely measurable maternal characteristics. Birth weight estimates using our equation are both prospectively derivable starting from the end of the second trimester and more accurate than any previously devised algorithms, including methods based upon ultrasonographic fetal biometric data.

Journal Article↗
PubMed1996

Predicting the use of prostheses by vascular amputees.

OBJECTIVE: To evaluate our accuracy in predicting the use of prostheses by patients undergoing major lower limb amputation. DESIGN: Prospective study, with multiple assessors, "blind" to the predictions made by each other. MATERIALS: Sixty-one patients (35 male: age 51-91, median 79) having their first major lower limb amputation. METHODS: Five members of the rehabilitation team (surgeon, specialist in prosthetics, nurse, physiotherapist and occupational therapist) each recorded predictions of prosthetic use and mobility before amputation and during the first 2 weeks thereafter. Patients were followed up 6-24 months later. RESULTS: At follow-up 17 patients had died. Of the remaining 44 (25 below-knee and 19 above-knee amputees), 23 of 27 (85%) who had been predicted as using prostheses were doing so, while only 11 of 17 (65%) had been correctly predicted as non-users. Nevertheless, only two of the patients not using prostheses contrary to prediction had ever had prostheses made for them, and both had developed problems with the other leg at a later date. Different members of the rehabilitation team were similar in their ability to predict outcome. CONCLUSIONS: Inappropriate fitting of prostheses can be kept to a minimum by a team approach to rehabilitation, but amputees may defy careful prediction by the development of new medical problems.

Aged↗
PubMed1997

Predicting length of stay in psychiatry.

BACKGROUND: Diagnostic Related Groups (DRGs) and Healthcare Resource Groups (HRGs) do not predict accurately length of stay or resources needed for treatment in psychiatry. This preliminary study assessed the relative contribution of severity of illness, in combination with other variables, in predicting length of stay. METHOD: Data were analysed on 115 consecutive admissions to a district psychiatric in-patient unit to assess the variables which most accurately predict length of stay. The variables included demographic data, diagnosis, clinical, social and behavioural measures. RESULTS: For initial admission, diagnosis of neurosis predicted shortest stay, but diagnosis alone accounted for only 14.6% of the variation in length of stay. Addition of Social Behaviour Scale score, living alone and specific psychiatric symptoms significantly increased the predictive value (adjusted R2 = 36.6%). Addition of variables available at discharge (use of ECT, major tranquillizers and antidepressants) significantly increased the adjusted R2 to 49.0%. Prediction of total length of hospitalization over a 12-month period, from the date of initial admission, indicated that mania predicted the longest stay and addition of other variables meant that only 18.9% of length of stay was predicted. CONCLUSION: If these results are borne out in a large study, they indicate that diagnostic or health related groups (DRGs) are only likely to be useful in psychiatry if they include more detailed social, clinical and behavioural variables.

Analysis of Variance↗
PubMed2000

A test of the model to predict unusually stable RNA hairpin loop stability.

To investigate the accuracy of a model [Giese et al., 1998, Biochemistry37:1094-1100 and Mathews et al., 1999, JMol Biol 288:911-940] that predicts the stability of RNA hairpin loops, optical melting studies were conducted on sets of hairpins previously determined to have unusually stable thermodynamic parameters. Included were the tetraloops GNRA and UNCG (where N is any nucleotide and R is a purine), hexaloops with UU first mismatches, and the hairpin loop of the iron responsive element, CAGUGC. The experimental values for the GNRA loops are in excellent agreement (deltaG degrees 37 within 0.2 kcal/mol and melting temperature (TM) within 4 degrees C) with the values predicted by the model. When the UNCG hairpin loops are treated as tetraloops, and a bonus of 0.8 kcal/mol included in the prediction to account for the extra stable first mismatch (UG), the measured and predicted values are also in good agreement (deltaG degrees 37 within 0.7 kcal/mol and TM within 3 degrees C). Six hairpins with unusually stable UU first mismatches also gave good agreement with the predictions (deltaG degrees 37 within 0.5 kcal/mol and TM within 8 degrees C), except for hairpins closed by wobble base pairs. For these hairpins, exclusion of the additional stabilization term for UU first mismatches improved the prediction (AG degrees 37 within 0.1 kcal/mol and TM within 3 degrees C). Hairpins with the iron-responsive element loop were not predicted well by the model, as measured deltaG degrees 37 values were at least 1 kcal/mol greater than predicted.

Base Pair Mismatch↗
PubMedDate not supplied

QSARs for 6-azasteroids as inhibitors of human type 1 5alpha-reductase: prediction of binding affinity and selectivity relative to 3-BHSD.

Quantitative structure-activity relationships (QSARs) are developed to describe the ability of 6-azasteroids to inhibit human type 1 5alpha-reductase. Models are generated using a set of 93 compounds with known binding affinities (K(i)) to 5alpha-reductase and 3beta-hydroxy-Delta(5)-steroid dehydrogenase/3-keto-Delta(5)-steroid isomerase (3-BHSD). QSARs are generated to predict K(i) values for inhibitors of 5alpha-reductase and to predict selectivity (S(i)) of compound binding to 3-BHSD relative to 5alpha-reductase. Log(K(i)) values range from -0.70 log units to 4.69 log units, and log(S(i)) values range from -3.00 log units to 3.84 log units. Topological, geometric, electronic, and polar surface descriptors are used to encode molecular structure. Information-rich subsets of descriptors are identified using evolutionary optimization procedures. Predictive models are generated using linear regression, computational neural networks (CNNs), principal components regression, and partial least squares. Compounds in an external prediction set are used for model validation. A 10-3-1 CNN is developed for prediction of binding affinity to 5alpha-reductase that produces root-mean-square error (RMSE) of 0.293 log units (R(2) = 0.97) for compounds in the external prediction set. Additionally, an 8-3-1 CNN is generated for prediction of inhibitor selectivity that produces RMSE = 0.513 log units (R(2) = 0.89) for the external prediction set. Models are further validated through Monte Carlo experiments in which models are generated after dependent variable values have been scrambled.

Azasteroids↗