Accurate prediction of HIV-1 drug response from the reverse transcriptase and protease amino acid sequences using sparse models created by convex optimization

MOTIVATION

Genotype-phenotype modeling problems are often overcomplete, or ill-posed, since the number of potential predictors-genes, proteins, mutations and their interactions-is large relative to the number of measured outcomes. Such datasets can still be used to train sparse parameter models that generalize accurately, by exerting a principle similar to Occam's Razor: When many possible theories can explain the observations, the most simple is most likely to be correct. We apply this philosophy to modeling the drug response of Type-1 Human Immunodeficiency Virus (HIV-1). Owing to the decreasing expense of genetic sequencing relative to in vitro phenotype testing, a statistical model that reliably predicts viral drug response from genetic data is an important tool in the selection of antiretroviral therapy (ART). The optimization techniques described will have application to many genotype-phenotype modeling problems for the purpose of enhancing clinical decisions.

RESULTS

We describe two regression techniques for predicting viral phenotype in response to ART from genetic sequence data. Both techniques employ convex optimization for the continuous subset selection of a sparse set of model parameters. The first technique, the least absolute shrinkage and selection operator, uses the l(1) norm loss function to create a sparse linear model; the second, the support vector machine with radial basis kernel functions, uses the epsilon-insensitive loss function to create a sparse non-linear model. The techniques are applied to predict the response of the HIV-1 virus to 10 reverse transcriptase inhibitor and 7 protease inhibitor drugs. The genetic data are derived from the HIV coding sequences for the reverse transcriptase and protease enzymes. When tested by cross-validation with actual laboratory measurements, these models predict drug response phenotype more accurately than models previously discussed in the literature, and other canonical techniques described here. Key features of the methods that enable this performance are the tendency to generate simple models where many of the parameters are zero, and the convexity of the cost function, which assures that we can find model parameters to globally minimize the cost function for a particular training dataset.

AVAILABILITY

Results, tables and figures are available at ftp://ftp.genesecurity.net.

SUPPLEMENTARY INFORMATION

An Appendix to accompany this article is available at Bioinformatics online.

Solid-phase/solution-phase combinatorial synthesis of neuroimmunophilin ligands

A novel solid-phase/solution-phase strategy for the synthesis of neuroimmunophilin ligands based on GPI 1046 was developed. The synthesis employs a solid-phase esterification strategy followed by a solution-phase pyruvic amide formation to produce multi-milligram quantities of discrete compounds for assay. The protocol was applied to a production library of 880 discrete compounds. A highlight of the strategy is an aqueous extractive purification of the final compounds using a novel liquid/ice extraction system developed for high throughput.

Historical perspective on lead biokinetic models

A historical review of the development of biokinetic model of lead is presented. Biokinetics is interpreted narrowly to mean only physiologic processes happening within the body. Proceeding chronologically, for each epoch, the measurements of lead in the body are presented along with mathematical models in an attempt to trace the convergence of observations from two disparate fields--occupational medicine and radiologic health--into some unified models. Kehoe's early balance studies and the use of radioactive lead tracers are presented. The 1960s saw the joint application of radioactive lead techniques and simple compartmental kinetic models used to establish the exchange rates and residence times of lead in body pools. The applications of stable isotopes to questions of the magnitudes of respired and ingested inputs required the development of a simple three-pool model. During the 1980s more elaborate models were developed. One of their key goals was the establishment of the dose-response relationship between exposure to lead and biologic precursors of adverse health effects.

The contribution of lead-contaminated house dust and residential soil to children's blood lead levels. A pooled analysis of 12 epidemiologic studies

In 1992, the U.S. Congress passed the Residential Lead-Based Paint Hazard Reduction Act, which requires the promulgation of health-based dust lead and soil lead standards for residential dwellings to prevent undue lead exposure in children. Unfortunately, the levels of lead in house dust and soil that are associated with elevated blood lead levels among U.S. children remain poorly defined. This pooled analysis was done to estimate the contributions of lead-contaminated house dust and soil to children's blood lead levels. The results of this pooled analysis, the most comprehensive existing epidemiologic analysis of childhood lead exposure, confirm that lead-contaminated house dust is the major source of lead exposure for children. These analyses further demonstrate that a strong relationship between interior dust lead loading and children's blood lead levels persists at dust lead levels considerably below the U.S. Department of Housing and Urban Development's current postabatement standards and the Environmental Protection Agency's guidance levels. Finally, these analyses demonstrate that a child's age, race, mouthing behaviors, and study-site specific factors influence the predicted blood lead level at a given level of exposure. These data can be used to estimate the potential health impact of alternative health-based lead standards for residential sources of lead exposure.

Bone lead as a biological marker in epidemiologic studies of chronic toxicity: conceptual paradigms

The skeleton contains the majority of the body's lead burden in both children and adults. The half-life of lead in bone is in the range of years to decades, depending on bone type, metabolic state, and subject age, among other things. Measurement of skeletal lead has benefited greatly from the recent development of X-ray fluorescence (XRF) instruments that can make rapid, safe, accurate, and relatively precise measurements of lead in bone. Two types of XRF technologies exist, LXRF and KXRF; this paper focuses on KXRF, which has been the most widely validated and used. KXRF is proving to be a powerful analytical methodology for evaluating bone lead levels as a measure of time-integrated (i.e., cumulative) lead dose in epidemiologic studies of the effects of chronic lead exposure. However, insufficient attention has been given to conceptualizing the paradigms by which bone lead levels reflect lead exposure and by which the skeleton serves as an endogenous source of lead. Consideration of these paradigms, which rely on bone lead kinetics, is necessary for the proper development of a priori hypotheses involving bone lead accumulation and release, the selection of bone sites for measurement by KXRF, and the design of epidemiologic studies involving bone lead dynamics. We discuss and present supporting evidence for a conceptual model that distinguishes two major paradigms of skeletal lead, including 1) bone lead as an indicator of cumulative lead exposure (bone lead as repository), and 2) bone lead as a source of body lead burden that is mobilizable into the circulation (bone lead as source). These two roles are not mutually exclusive. Instead, they are components of the processes controlling lead accumulation into and release from bone over time. Developing successful strategies for distinguishing these two processes in epidemiologic studies will require separate measurements of lead in cortical and trabecular bone and additional measurement of specific markers of bone mineral turnover and resorption. It may also involve developing accurate methods for evaluating lead in labile compartments of the circulation, such as plasma, as a potentially useful and responsive measure of bone lead release, of the partitioning of circulatory lead, and of the toxicological significance of lead released from bone to other target organs.

Models of delayed recovery

After a biologic insult has impaired function of the developing central nervous system, recovery may not become apparent for years. Probability models adopted from the carcinogenesis, developmental neurobiology, learning decay, and stochastic process literatures are presented so that assumptions about apparent delays in the recovery process can be tested with data from longitudinal studies after a temporally circumscribed adverse event/exposure. This process of evaluating multiple models is exemplified with one data set. Nonlinear models of recovery are important because some children with early deficits first show improvement months to years later.

Psychopathology and cognitive decline in dementia

This study investigated the correlation between the degree and quality of cognitive impairment and the presence and degree of severity of psychopathology. Sixty-four demented outpatients were rated for cognition by the Mental Status Questionnaire and for psychopathology by the BEHAVE-AD Reisberg questionnaire. A statistical analysis of the correlations between the cognitive and psychopathological parameters was performed. The study indicated a high rate of psychopathology (as high as 83% of the patients). No significant correlation was found between the cognitive grading score and the presence of psychopathology, except in the space of declined spatial orientation and the occurrence of hallucinations. Psychopathology is a common phenomenon occurring in the course of dementia in the form of Alzheimer's disease or multi-infarct dementia. There is a certain increase in the degree of psychopathology that correlated with cognitive impairment. The high correlation between impaired spatial orientation and the presence of hallucinations seems to point to a common neuropathological substrate.

Pre- and postnatal lead exposure and behavior problems in school-aged children

The association between early lead exposure and later problem behaviors was evaluated prospectively in a cohort of 8-year-old children born during a 12-month period at one hospital. Lead levels in umbilical cord blood (means = 6.8 micrograms/dl, SD = 3.1) and the dentin of a shed deciduous tooth (means = 3.4 micrograms/g, SD = 2.4) provided measures of prenatal and postnatal exposure, respectively. Ratings on the Teacher Report Form of the Child Behavior Profile provided information about children's problem behaviors. Cord blood lead level was not associated with the overall prevalence or nature of problem behaviors. In both crude and adjusted analyses, tooth lead level was significantly associated with total problem behavior scores (approximately 2 points in T score per log unit increase in tooth lead). Significant tooth lead-associated increases in both internalizing and externalizing scores were also observed (approximately 1.5 points in T score per log unit increase). Weaker associations were noted between tooth lead level and the prevalence of "extreme" problem behavior scores. The extent to which these associations reflect residual confounding is uncertain. These data suggest, however, that social and emotional dysfunctions are correlates and may be expressions of increased lead exposure.

Age-related differences in speed of processing: unconfounding age and experience

This study was designed to explore age constraints on speed of processing on a lexical decision task. In order to unconfound age and experience, the participants were presented with two lexical decision tasks, one in German (the native language) and one in English (the second language), as well as a symbol matching task. Three groups of subjects were formed: (1) 16-year-olds who had received formal instruction in English for 5 years, (2) 16-year-olds who had received only 1 year of instruction in English, and (3) 14-year-olds who had received 3 years of instruction in English. Inclusion of these three groups permitted the study of the effects of language experience in the absence of the usual age-experience confound. When the lexical decision task involved German words and nonwords, the older children responded more quickly than the younger subjects. However, when the stimulus items were English words and non-words, this age-related progression was disrupted and response speed was related to experience with English as a second language. These results suggest that experience is an important factor to consider when trying to account for lexical access times. Implications for understanding age-related differences are discussed.

Dentin as a possible bio-epidemiological measure of exposure to mercury

The subtle human health effects from prolonged exposure to small amounts of mercury vapor are unknown. It has been difficult to study possible effects of low-dose exposure for lack of a good measure of long-term exposure. Current methods which use blood, urine, hair, and nails reliably measure only recent exposures. Long-term exposure to lead has been measured using levels found in human dentin. The purpose of this study was to determine whether mercury also accumulates in dentin. In this study, dentin from 16 human teeth, all without dental amalgam restorations, was analyzed by thermal neutron activation analysis. The teeth were selected from people with and without dental mercury amalgam restorations elsewhere in their dentitions. Mercury was found in amounts up to 5.9 ppm. While the highest mercury level was from a sample from someone who had dental amalgam restorations elsewhere in their dentition, the second highest was from someone who had no amalgam restorations. Also, a sample which was not used for any statistical comparisons but which was analyzed because the tooth contained an amalgam restoration had one of the lowest levels of mercury. These results, while inconclusive due to a small sample size, suggest that inorganic mercury vapor is a relatively small contributor to the overall body burden of mercury.

Uptake of serotonin into platelets of senile dementia of the Alzheimer type patients

[3H]-serotonin uptake in platelets was studied in 26 patients with senile dementia of the Alzheimer type (SDAT), 29 age-matched normal elderly persons, and 21 young subjects. The results showed a significantly lower uptake of serotonin into platelets of SDAT patients than in those of elderly controls (p < .02). Uptake of serotonin into platelets of elderly normal subjects was significantly lower than in the young (p < .01). These differences were due to reduced Vmax, whereas Km was unchanged. Isolated plasma from SDAT patients did not affect the uptake of serotonin into platelets of elderly controls, and vice versa, i.e., plasma obtained from elderly controls did not affect serotonin uptake in SDAT patients. The results indicate that serotonin uptake into platelets is reduced in normal aging, and more so in SDAT. Moreover, the reduced uptake in SDAT is not caused by a plasma factor in SDAT patients. No correlation was found between serotonin uptake and degree of cognitive impairment in SDAT patients.

Pre- and postnatal low-level lead exposure and children's dysfunction in school

The contributions of pre- and postnatal low-level lead exposures to the risk of learning problems were evaluated among 1923 children who were born in one Boston hospital in 1979-1980 and followed to age 8 years. In this relatively privileged group, more than 20% of the children had a mother with some formal postgraduate education. Prenatal lead exposure was estimated with a measurement of umbilical cord blood lead content, and postnatal lead exposure was approximated with measurement of lead in the dentin of an exfoliated deciduous tooth. Information about potential confounders and effect modifiers was obtained from maternal interview shortly after delivery and from a mailed questionnaire completed and returned when the child was approximately 6 years old. An assessment of each child's function in school was provided by the teacher, who completed a questionnaire near the end of the school year in which the child reached the age of 8 years. We considered a learning problem to be related to lead exposure if its adjusted prevalence increased with each loge increase in lead, and if the adjusted prevalence was elevated among children with high levels (i.e., approximating the highest decile) of umbilical cord blood lead (i.e., > or = 10 micrograms/dl) or dentin lead (i.e., > or = 5 micrograms/g). Girls with elevated umbilical cord blood lead levels were more likely than their peers to be dependent and inpersistent and to display an inflexible and inappropriate approach to tasks (defined as the "tasks" cluster). Boys with elevated umbilical cord blood lead levels were more likely than others to have difficulty with both simple directions and sequences of directions. Among girls, elevated deciduous tooth dentin lead content was associated with reading and spelling difficulties, the tasks cluster, and with "not functioning as well as peers." Elevated dentin lead levels were not overrepresented among boys with any of the assessed learning clusters. These findings are consistent with the inference that lead levels still prevalent among children (i.e., blood < 15 micrograms/dl) are associated with some learning problems in girls.

Diagnostic reasoning of high- and low-domain-knowledge clinicians: a reanalysis

Thinking-aloud protocols provided by Joseph and Patel were reanalyzed to determine the extent to which their conclusions could be replicated by independently developed coding schemes. The data set consisted of protocols from four cardiologists (low domain knowledge = LDK) and four endocrinologists (high domain knowledge = HDK), individually working on a diagnostic problem in endocrinology. The two analyses agree that the HDK physicians related data to potential diagnoses more than did the LDK group and were more focused on the correct diagnostic components. However, the reanalysis found no meaningful difference between the groups in diagnostic accuracy, speed of diagnosis, or the breadth of the search space used to seek a solution. In the reanalysis, the HDK physicians employed more single-cue inference and less multiple-cue inference. The generalizability of results of protocol-analysis studies can be assessed by using several complementary coding schemes.

Antecedents and correlates of hypoplastic enamel defects of primary incisors

Four hundred and fifty-five exfoliated primary incisors were obtained from children whose mothers provided information about a wide range of pre-, peri-, and postnatal characteristics of both the mother and child. These teeth then were examined for the presence of hypoplastic enamel defects (HED). The basic form of the null hypothesis tested was that children who had HED of a primary incisor did not differ from those who did not have such a defect. Of the primary incisors examined, 18.5% had HED (25.0% maxillary and 10.1% mandibular). The following items were found to be associated most strongly (P less than 0.003) with an increase in a child's risk of developing HED; 1) maternal antenatal history of smoking, higher prepregnancy weight, and failure to obtain prenatal care during the first trimester; 2) prematurity, low birth-weight and their associated correlates; and 3) postnatal measles infection. Left-handedness, maternal tea and Tylenol (McNeil Consumer Products Co., Fort Washington, PA) consumption, and failure to screen for undue lead burden were associated less prominently (P less than 0.05) with HED prevalence. Season of birth and serum and dentin lead levels were not related to the prevalence of HED. Many of these risk factors are also covariates of low socioeconomic status such as suboptimal nutrition and increased risk of infection. Additional investigation is needed to delineate the associations between specific pre- and perinatal nutritional and infectious factors, socioeconomic status, and HED development.

Low-level lead exposure and children's cognitive function in the preschool years

In a cohort of 170 middle and upper-middle class children participating in a prospective study of child development and low-level lead exposure, higher blood lead levels at age 24 months were associated with lower scores at age 57 months on the McCarthy Scales of Children's Abilities. The mean blood lead level at age 24 months was 6.8 micrograms/dL (SD = 6.3; 75th, 90th, and 99th percentiles: 8.8, 13.7, 23.6, respectively) and for all but 1 child was less than 25 micrograms/dL, the current definition of an "elevated" level. After adjustment for confounding, scores on the General Cognitive Index decreased approximately 3 points (SE = 1.4) for each natural log unit increase in 24-month blood lead level. The inverse association between lead level and performance was especially prominent for visual-spatial and visual-motor integration skills. Higher prenatal exposures were not associated with lower scores at 57 months except in the subgroup of children with "high" concurrent blood lead levels (ie, greater than or equal to 10 micrograms/dL). The concentration of lead in the dentine of shed deciduous teeth was not significantly associated with children's performance after adjustment for confounding.

Avoiding blood transfusion. A report of two cases

It is prudent to avoid blood transfusion whenever possible. Recent reports question the standards previously considered acceptable for blood transfusion. Acute hemorrhage followed by hypovolemia, tachycardia and orthostatic hypotension with dizziness are still considered acceptable criteria for automatic blood transfusion in most patients. We report two cases in which hypovolemia, tachycardia and orthostatic hypotension with dizziness were present. Transfusion was avoided. Hypovolemia, tachycardia and orthostatic hypotension with dizziness do not require transfusion in a few carefully selected and well-monitored patients.

Double valve replacement for lupus valvulitis: report of a case and review of the literature

Hemodynamically significant lupus valvulitis, requiring valve replacement, is rare: 21 cases have been reported so far in the literature, and only 2 of these have involved double valve replacement. We describe an additional case of double valve replacement in a patient with systemic lupus erythematosus. The histopathologic and clinical features of this case suggest that valvular involvement resulted from both acute and chronic disease processes. Medical success in the treatment of systemic lupus erythematosus, especially that achieved through prolonged or high-dose steroid therapy, may cause chronic valvular disease to become a more common surgical problem. A review of the literature supports this contention.

Alzheimer's dementia and binding to alpha 2 adrenoreceptors in platelets

Seventy-five patients with probable Alzheimer's disease were screened for binding of alpha 2 receptors (A2R) to their platelet membranes; the results were compared with 51 age- and sex-matched controls. Receptor binding assays were performed using [3H] Yohimbine as the radioligand. The results showed a higher binding capacity in the demented population as compared to the control group (2.18 +/- 0.15 fmol/mg protein, as compared to 1.73 +/- 0.13, P less than 0.03). This increased binding to platelets in the demented patients was more prominent in demented females: 34% higher binding as compared with female controls (2.06 +/- 0.5 vs 1.54 +/- 0.04). The difference between demented and normal males was less (2.34 +/- 0.05 vs 1.88 +/- 0.05). The results indicate an involvement of the A2R system, either primarily or secondarily, in the disease process. Since there is an overlap between results from the patients with Alzheimer's disease and the normal subjects, A2R may serve as only a supportive marker for Alzheimer's disease.

Low-level lead exposure, social class, and infant development

A prospective cohort study was conducted to assess the association between early development and low-level prenatal and postnatal lead exposure. Infants' performance between 6 and 24 months on the Mental Development Index of the Bayley Scales of Infant Development declined with increasing concentration of lead in blood, but the decline varied with children's age at exposure, level of exposure, and socioeconomic status. Within the second year of life, the performance of children in lower socioeconomic strata was adversely affected at lower levels of prenatal exposure (blood lead levels of 6 to 7 micrograms/dl) than was the performance of children in higher socioeconomic strata. However, even the performance of these advantaged infants was lower when cord blood lead level exceeded 10 micrograms/dl, well below the figure currently regarded as the maximum permissable level for young children. Exploratory analyses suggested that early postnatal blood lead levels between 10 and 25 micrograms/dl were also associated with lower Mental Development Index scores, but only among children in lower socioeconomic strata.