Merits of random forests emerge in evaluation of chemometric classifiers by external validation.

Real-world applications will inevitably entail divergence between samples on which chemometric classifiers are trained and the unknowns requiring classification. This has long been recognized, but there is a shortage of empirical studies on which classifiers perform best in 'external validation' (EV), where the unknown samples are subject to sources of variation relative to the population used to train the classifier. Survey of 286 classification studies in analytical chemistry found only 6.6% that stated elements of variance between training and test samples. Instead, most tested classifiers using hold-outs or resampling (usually cross-validation) from the same population used in training. The present study evaluated a wide range of classifiers on NMR and mass spectra of plant and food materials, from four projects with different data properties (e.g., different numbers and prevalence of classes) and classification objectives. Use of cross-validation was found to be optimistic relative to EV on samples of different provenance to the training set (e.g., different genotypes, different growth conditions, different seasons of crop harvest). For classifier evaluations across the diverse tasks, we used ranks-based non-parametric comparisons, and permutation-based significance tests. Although latent variable methods (e.g., PLSDA) were used in 64% of the surveyed papers, they were among the less successful classifiers in EV, and orthogonal signal correction was counterproductive. Instead, the best EV performances were obtained with machine learning schemes that coped with the high dimensionality (914-1898 features). Random forests confirmed their resilience to high dimensionality, as best overall performers on the full data, despite being used in only 4.5% of the surveyed papers. Most other machine learning classifiers were improved by a feature selection filter (ReliefF), but still did not out-perform random forests.

Using a logical model to predict the growth of yeast.

BACKGROUND: A logical model of the known metabolic processes in S. cerevisiae was constructed from iFF708, an existing Flux Balance Analysis (FBA) model, and augmented with information from the KEGG online pathway database. The use of predicate logic as the knowledge representation for modelling enables an explicit representation of the structure of the metabolic network, and enables logical inference techniques to be used for model identification/improvement. RESULTS: Compared to the FBA model, the logical model has information on an additional 263 putative genes and 247 additional reactions. The correctness of this model was evaluated by comparison with iND750 (an updated FBA model closely related to iFF708) by evaluating the performance of both models on predicting empirical minimal medium growth data/essential gene listings. CONCLUSION: ROC analysis and other statistical studies revealed that use of the simpler logical form and larger coverage results in no significant degradation of performance compared to iND750.

Functional bioinformatics for Arabidopsis thaliana.

MOTIVATION: The genome of Arabidopsis thaliana, which has the best understood plant genome, still has approximately one-third of its genes with no functional annotation at all from either MIPS or TAIR. We have applied our Data Mining Prediction (DMP) method to the problem of predicting the functional classes of these protein sequences. This method is based on using a hybrid machine-learning/data-mining method to identify patterns in the bioinformatic data about sequences that are predictive of function. We use data about sequence, predicted secondary structure, predicted structural domain, InterPro patterns, sequence similarity profile and expressions data. RESULTS: We predicted the functional class of a high percentage of the Arabidopsis genes with currently unknown function. These predictions are interpretable and have good test accuracies. We describe in detail seven of the rules produced.

Predicting gene function in Saccharomyces cerevisiae.

MOTIVATION: S.cerevisiae is one of the most important model organisms, and has has been the focus of over a century of study. In spite of these efforts, 40% of its open reading frames (ORFs) remain classified as having unknown function (MIPS: Munich Information Center for Protein Sequences). We wished to make predictions for the function of these ORFs using data mining, as we have previously successfully done for the genomes of M.tuberculosis and E.coli. Applying this approach to the larger and eukaryotic S.cerevisiae genome involves modifying the machine learning and data mining algorithms, as this is a larger organism with more data available, and a more challenging functional classification. RESULTS: Novel extensions to the machine learning and data mining algorithms have been devised in order to deal with the challenges. Accurate rules have been learned and predictions have been made for many of the ORFs whose function is currently unknown. The rules are informative, agree with known biology and allow for scientific discovery. AVAILABILITY: All predictions are freely available from http://www.genepredictions.org, all datasets used in this study are freely available from http://www.aber.ac.uk/compsci/Research/bio/dss/yeastdataand software for relational data mining is available from http://www.aber.ac.uk/compsci/Research/bio/dss/polyfarm.

The utility of different representations of protein sequence for predicting functional class.

MOTIVATION: Data Mining Prediction (DMP) is a novel approach to predicting protein functional class from sequence. DMP works even in the absence of a homologous protein of known function. We investigate the utility of different ways of representing protein sequence in DMP (residue frequencies, phylogeny, predicted structure) using the Escherichia coli genome as a model. RESULTS: Using the different representations DMP learnt prediction rules that were more accurate than default at every level of function using every type of representation. The most effective way to represent sequence was using phylogeny (75% accuracy and 13% coverage of unassigned ORFs at the most general level of function: 69% accuracy and 7% coverage at the most detailed). We tested different methods for combining predictions from the different types of representation. These improved both the accuracy and coverage of predictions, e.g. 40% of all unassigned ORFs could be predicted at an estimated accuracy of 60% and 5% of unassigned ORFs could be predicted at an estimated accuracy of 86%.

Extracellular calcium depletion as a mechanism of short-term synaptic depression.

Recent experiments have demonstrated that normal neural activity can cause significant decrements in external calcium levels, and that these decrements mediate a form of short-term synaptic depression. These findings raise the possibility that certain forms of short-term synaptic depression at glutamatergic synapses throughout the mammalian CNS may be influenced by similar changes in external calcium. We use a computational model of the extracellular space, combined with experimental data on calcium consumption, to show that such short-term depression can be accounted for by changes in calcium just outside active synapses, provided that external calcium diffusion is restricted. Remarkably, the model suggests the novel possibility that synapses may possess private pools of external calcium that enforce some forms of short-term depression in a synapse-specific manner.

Fate of postharvest-applied dichlorvos in stored and processed dates.

The fate of dichlorvos (DDVP) on dates after storage and processing of postharvest-treated fruits has been investigated. Residues were determined using GC-ECD after extraction of the fruits. A postharvest application was made to fruits at different stages of maturity: khalal fruits (mature full colored stage), rutab fruits (soft brown stage), and the mature tamer fruits (hard raisin-like stage). The fate of the residues was followed during the storage at refrigerated, room, and summer average temperatures (3, 22, and 43 degrees C). The amount of residues absorbed varied with the level of maturity. The rate of the loss of the residues was found to follow first-order kinetics. First-order rate constants were calculated for the different levels of maturity. The rate of the loss of DDVP increased as the temperature of storage increased. Also, it decreased with increasing maturation of the dates, which was characterized by a decrease in moisture content as well as water activity and an increase in sugar content. The period of storage study was limited by the time of maturation to the next stage. Most common home-cooking methods, including dehydration, jam-making, and syrup-making, resulted in significant decreases in residue levels. Only 0-13% of initial DDVP residues was detected in final products.

Description of a growth simulation model for predicting the effect of diet on broiler composition and growth.

The growth simulation program, BPHL (Bromley Park Hatcheries Limited), is a computerized, mechanistic, deterministic and dynamic approach to the evaluation of the effects of diet on broiler carcass composition and growth. Daily growth is simulated with information on the initial age and live weight of the bird, number of days over which the diet is to be fed, protein and amino acid densities of the diet, dietary metabolizable energy, and whether feed intake is to be simulated or data provided. Output provides information on a daily basis with respect to daily and accumulated deposition and current bird status for protein, fat, water, and ash body content. Carcass weight, feather weight, live weight, feed eaten, feed deprivation heat loss, limiting amino acids, food conversion ratio, and percentage carcass fat are also provided. The approach employs empirically derived first-limiting amino acid coefficients relating to accretion efficiency and dietary concentration to define limits of protein retention, uses mathematical expressions describing feed intake and heat loss trajectories as datum patterns prescriptive of the strain, introduces calibration as a device for improving correspondence between simulated and field performance, and relies on an assumption that deviations to the datum patterns of food intake and heat output caused by strain and environmental factors can be duplicated by simple multiplers acting on the expressions. The program may be used as a tool for exploring the predicted effect of specific dietary characteristics and strain parameters on growth, body composition, and performance.

Warmr: a data mining tool for chemical data.

Data mining techniques are becoming increasingly important in chemistry as databases become too large to examine manually. Data mining methods from the field of Inductive Logic Programming (ILP) have potential advantages for structural chemical data. In this paper we present Warmr, the first ILP data mining algorithm to be applied to chemoinformatic data. We illustrate the value of Warmr by applying it to a well studied database of chemical compounds tested for carcinogenicity in rodents. Data mining was used to find all frequent substructures in the database, and knowledge of these frequent substructures is shown to add value to the database. One use of the frequent substructures was to convert them into probabilistic prediction rules relating compound description to carcinogenesis. These rules were found to be accurate on test data, and to give some insight into the relationship between structure and activity in carcinogenesis. The substructures were also used to prove that there existed no accurate rule, based purely on atom-bond substructure with less than seven conditions, that could predict carcinogenicity. This results put a lower bound on the complexity of the relationship between chemical structure and carcinogenicity. Only by using a data mining algorithm, and by doing a complete search, is it possible to prove such a result. Finally the frequent substructures were shown to add value by increasing the accuracy of statistical and machine learning programs that were trained to predict chemical carcinogenicity. We conclude that Warmr, and ILP data mining methods generally, are an important new tool for analysing chemical databases.

Accurate prediction of protein functional class from sequence in the Mycobacterium tuberculosis and Escherichia coli genomes using data mining.

The analysis of genomics data needs to become as automated as its generation. Here we present a novel data-mining approach to predicting protein functional class from sequence. This method is based on a combination of inductive logic programming clustering and rule learning. We demonstrate the effectiveness of this approach on the M. tuberculosis and E. coli genomes, and identify biologically interpretable rules which predict protein functional class from information only available from the sequence. These rules predict 65% of the ORFs with no assigned function in M. tuberculosis and 24% of those in E. coli, with an estimated accuracy of 60-80% (depending on the level of functional assignment). The rules are founded on a combination of detection of remote homology, convergent evolution and horizontal gene transfer. We identify rules that predict protein functional class even in the absence of detectable sequence or structural homology. These rules give insight into the evolutionary history of M. tuberculosis and E. coli.

The co-occurrence of psychiatric and substance use diagnoses in adolescents in different service systems: frequency, recognition, cost, and outcomes.

The frequency, severity, recognition, cost, and outcomes of adolescent substance use comorbidity were analyzed in the Fort Bragg Demonstration Project. Comorbidity was defined as the co-occurrence of substance use disorder (SUD) with other psychiatric diagnosis. The sample consisted of 428 adolescent clients whose providers' diagnoses were compared with research diagnoses. The project identified 59 clients (13.8%) with SUD, all with additional psychiatric diagnoses. Providers recognized only 21 of these 59 comorbid cases. The frequency and severity of comorbidity did not differ between service system samples, although recognition did. Comorbid clients had more behavior problems and more functioning impairment, and their average treatment cost ($29,057) was more than twice as high as that of noncomorbid clients ($13,067). Mental health outcomes were not influenced by type of service system, comorbid diagnosis, or treatment. Screening for and prevention of SUD are discussed as a potential cost-savings opportunity in mental health services.

Cascaded multiple classifiers for secondary structure prediction.

We describe a new classifier for protein secondary structure prediction that is formed by cascading together different types of classifiers using neural networks and linear discrimination. The new classifier achieves an accuracy of 76.7% (assessed by a rigorous full Jack-knife procedure) on a new nonredundant dataset of 496 nonhomologous sequences (obtained from G.J. Barton and J.A. Cuff). This database was especially designed to train and test protein secondary structure prediction methods, and it uses a more stringent definition of homologous sequence than in previous studies. We show that it is possible to design classifiers that can highly discriminate the three classes (H, E, C) with an accuracy of up to 78% for beta-strands, using only a local window and resampling techniques. This indicates that the importance of long-range interactions for the prediction of beta-strands has been probably previously overestimated.

Dendritic spikes and their influence on extracellular calcium signaling.

Extracellular calcium is critical for many neural functions, including neurotransmission, cell adhesion, and neural plasticity. Experiments have shown that normal neural activity is associated with changes in extracellular calcium, which has motivated recent computational work that employs such fluctuations in an information-bearing role. This possibility suggests that a new style of computing is taking place in the mammalian brain in addition to current 'circuit' models that use only neurons and connections. Previous computational models of rapid external calcium changes used only rough approximations of calcium channel dynamics to compute the expected calcium decrements in the extracellular space. Using realistic calcium channel models, experimentally measured back-propagating action potentials, and a model of the extracellular space, we computed the fluctuations in external calcium that accrue during neural activity. In this realistic setting, we showed that rapid, significant changes in local external calcium can occur when dendrites are invaded by back-propagating spikes, even in the presence of an extracellular calcium buffer. We further showed how different geometric arrangements of calcium channels or dendrites prolong or amplify these fluctuations. Finally, we computed the influence of experimentally measured synaptic input on peridendritic calcium fluctuations. Remarkably, appropriately timed synaptic input can amplify significantly the decrement in external calcium. The model shows that the extracellular space and the calcium channels that access it provide a medium that naturally integrates coincident spike activity from different dendrites that intersect the same tissue volume.

On the optimization of classes for the assignment of unidentified reading frames in functional genomics programmes: the need for machine learning.

At present, the assignment of function to novel genes uncovered by the systematic genome-sequencing programmes is a problem. Many studies anticipate that this can be achieved by analysing patterns of gene expression via the transcriptome, proteome and metabolome. Thus, functional genomics is, in part, an exercise in pattern classification. Because many genes have known functional classes, the problem of predicting their functional class is a supervised learning problem. However, most pattern classification methods that have been applied to the problem have been unsupervised clustering methods. Consequently, the best classification tools have not always been used. Furthermore, the present functional classes are suboptimal and new unsupervised clustering methods are needed to improve them. Better-structured functional classes will facilitate the prediction of biochemically testable functions.

Is it better to combine predictions?

We have compared the accuracy of the individual protein secondary structure prediction methods: PHD, DSC, NNSSP and Predator against the accuracy obtained by combing the predictions of the methods. A range of ways of combing predictions were tested: voting, biased voting, linear discrimination, neural networks and decision trees. The combined methods that involve 'learning' (the non-voting methods) were trained using a set of 496 non-homologous domains; this dataset was biased as some of the secondary structure prediction methods had used them for training. We used two independent test sets to compare predictions: the first consisted of 17 non-homologous domains from CASP3 (Third Community Wide Experiment on the Critical Assessment of Techniques for Protein Structure Prediction); the second set consisted of 405 domains that were selected in the same way as the training set, and were non-homologous to each other and the training set. On both test datasets the most accurate individual method was NNSSP, then PHD, DSC and the least accurate was Predator; however, it was not possible to conclusively show a significant difference between the individual methods. Comparing the accuracy of the single methods with that obtained by combing predictions it was found that it was better to use a combination of predictions. On both test datasets it was possible to obtain a approximately 3% improvement in accuracy by combing predictions. In most cases the combined methods were statistically significantly better (at P = 0.05 on the CASP3 test set, and P = 0.01 on the EBI test set). On the CASP3 test dataset there was no significant difference in accuracy between any of the combined method of prediction: on the EBI test dataset, linear discrimination and neural networks significantly outperformed voting techniques. We conclude that it is better to combine predictions.

Prediction of the soluble myoglobin content of cooked burgers.

The time-temperature profiles for cooking in-house made beef and lamb burgers were determined using a thermocouple placed in the centre of the burger. From these data the soluble myoglobin remaining in the burger was predicted using kinetic data from previously reported model experiments. First order kinetics were assumed for the denaturation of the myoglobin. A good correlation between observed and predicted data was observed. Thus the "degree of doneness"of different meats can be predicted when cooked under specified conditions.

Effect of muscle type, salt and pH on cooked meat haemoprotein formation in lamb and beef.

The rate of cooked meat haemoprotein formation, measured as the rate of loss of myoglobin solubility, in lamb was dependent on the muscles anatomical location and temperature. Lamb longissimus dorsi musle at 55 to 70°C formed cooked meat haemoprotein more rapidly than the muscles in the shoulder and leg. The formation in lamb was more rapid than in beef. The rate in high pH beef (7.25) l. dorsi was lower than found in beef l. dorsi of normal pH but in low pH lamb (5.38) l. dorsi the rate was, at most temperatures, also slower than found in this muscle from lamb of normal pH. In the presence of NaCl the rate of cooked meat haemoprotein formation was faster (almost doubled at 2g/100g meat) than found in the corresponding salt free lamb and beef samples. Other additives commonly added to meat products (mechanically recovered meat, oil, polyphosphates, soya, whey and caseinate) had little effect on the rate of cooked meat haemoprotein formation, at the levels normally used in meat products. It is concluded that for lamb products little if any myoglobin will remain soluble, and the products will look cooked before the recommended thermal treatment to inactivate Escherichia coli O157:H7 has been achieved. ©

Linear model of nitrogen balance and examination of the nature of true metabolisable energy and its nitrogen corrected form.

1. The nature of nitrogen (N) corrected true metabolisable energy (TMEN) was derived using a linear model of N balance, constructed from the relationship between excreted and ingested N. 2. TME was described in terms of a regression line, formed from 'fed' points relating energy voided to energy ingested (GE), as GE - (afed + bGE) + afast. On assignment of theoretical excreta and ingested energy components, a deviation from conceptual metabolisable energy (MEc), equal to the difference between afed and afast, was established and attributed to metabolic urinary energy (UmE). 3. The N balance model is based on the form of relationship between N excreted and N ingested (NI) that exhibits a linear deviation at 'initial' rates of N ingestion. The model postulates the following: The deviation is the result of a sparing effect of ingested N on the N component of UmE, viz. metabolic urinary N (UmN); The magnitude of UmN, through 'initial' values of fed N, is described by an intercept component, aNp, and a slope quantity, -(bNr - bNna) NI, where bNna and bNr are respectively the slopes of N excretion through 'initial' and 'subsequent' rates of ingested food N; The magnitude of the deviation from zero nitrogen balance (ZNB) through 'initial' and 'subsequent' rates of ingested N is the sum of the previous terms and aNm - (1 - bNr) NI, where aNm is the intercept component representing maintenance losses of N at fasting and (1-bNr) NI is the quantity of fed N retained to replace maintenance N loss. 4. Application of the appropriate energetic forms of UmN and aNm, viz. Et aNp - Et (bNr - bNna) NI and EuaNm, to the expression for obtaining TME, demonstrated that TME exceeded MEc by the quantities Et (bNr - bNna) NI and Et aNp, for test food intakes resulting in 'initial' and 'subsequent' rates of food N, respectively. 5. Application of appropriate energetic components of the model to simulate correction of TME to ZNB, demonstrated TMEZNB to be a biased quantity, deviating from MEc by the amount -Eu (1 - bNr) NI or expressed as an excreta energy slope component, [formula: see text] where Eu is an appropriate energy coefficient. An alternative perspective is that ZNB correction removes the energetic form of UmN as a source of bias, but introduces one related to EuaNm. Its nature may be perceived by regarding TME as a function of a regression line relating energy excreted (EE) to energy ingested that has been corrected for UmN energetic bias and is pivoting on a fulcrum vertically aligned with the position of ZNB on the GE (x) axis. The regression line rotates anti-clockwise in response to ZNB correction by an amount equal to the magnitude of EuaNm measured on the EE (y) axis from the point of interception. 6. The study identified processes that may be employed to remove bias and improve precision of TME.