Computers in chemistryComputers in chemistry V 0380 Active Learning with Support Vector Machines in the Drug Discovery Process. -(WARMUTH*, M. K.; LIAO, J.; RAETSCH, G.; MATHIESON, M.; PUTTA, S.; LEMMEN, C.; J. Chem. Inf. Comput. Sci. 43 (2003) 2, 667-673; Comp. Sci. Dep., Univ. Calif., Santa Cruz, CA 95064, USA; Eng.) -Lindner 22-232

We investigate the following data mining problem from computer-aided drug design: From a large collection of compounds, find those that bind to a target molecule in as few iterations of biochemical testing as possible. In each iteration a comparatively small batch of compounds is screened for binding activity toward this target. We employed the so-called "active learning paradigm" from Machine Learning for selecting the successive batches. Our main selection strategy is based on the maximum margin hyperplane-generated by "Support Vector Machines". This hyperplane separates the current set of active from the inactive compounds and has the largest possible distance from any labeled compound. We perform a thorough comparative study of various other selection strategies on data sets provided by DuPont Pharmaceuticals and show that the strategies based on the maximum margin hyperplane clearly outperform the simpler ones.

Purpose Complete response to induction chemotherapy is observed in ~60% of patients with newly diagnosed non-M3 acute myelogenous leukemia (AML). However, no methods exist to predict with high accuracy at the individual patient level the response to standard AML induction therapy. Experimental Design We applied single-cell network profiling (SCNP) using flow cytometry, a tool that allows a comprehensive functional assessment of intracellular signaling pathways in heterogeneous tissues, to two training cohorts of AML samples (n = 34 and 88) to predict the likelihood of response to induction chemotherapy. Results In the first study, univariate analysis identified multiple signaling “nodes” (readouts of modulated intracellular signaling proteins) that correlated with response (i.e., AUCROC ≥ 0.66; P ≤ 0.05) at a level greater than age. After accounting for age, similar findings were observed in the second study. For patients <60 years old, complete response was associated with the presence of intact apoptotic pathways. In patients ≥60 years old, nonresponse was associated with FLT3 ligand–mediated increase in phosphorylated Akt and phosphorylated extracellular signal-regulated kinase. Results were independent of cytogenetics, FLT3 mutational status, and diagnosis of secondary AML. Conclusions These data emphasize the value of performing quantitative SCNP under modulated conditions as a basis for the development of tests highly predictive for response to induction chemotherapy. SCNP provides information distinct from other known prognostic factors such as age, secondary AML, cytogenetics, and molecular alterations and is potentially combinable with the latter to improve clinical decision making. Independent validation studies are warranted.

The performance of docking studies into protein active sites constructed by homology model building was investigated using CDK2 and factor VIIa screening data sets. When the sequence identity between model and template near the binding site area is greater than approximately 50%, roughly 5 times more active compounds are identified than would be found randomly. This performance is comparable to docking to crystal structures.

The shape of and the chemical features of a ligand are both critical for biological activity. This paper presents a strategy that uses these descriptors to build a computational model for virtual screening of bioactive compounds. Molecules are represented in a binary shape-feature descriptor space as bit-strings, and their relative activities are used to identify the subset of the bit-string that is most relevant to bioactivity. This subset is used to score virtual libraries. We describe the computational details of the method and present an example validation experiment on thrombin inhibitors.

A greater understanding of the function of the human immune system at the single-cell level in healthy individuals is critical for discerning aberrant cellular behavior that occurs in settings such as autoimmunity, immunosenescence, and cancer. To achieve this goal, a systems-level approach capable of capturing the response of the interdependent immune cell types to external stimuli is required. In this study, an extensive characterization of signaling responses in multiple immune cell subpopulations within PBMCs from a cohort of 60 healthy donors was performed using single-cell network profiling (SCNP). SCNP is a multiparametric flow cytometry-based approach that enables the simultaneous measurement of basal and evoked signaling in multiple cell subsets within heterogeneous populations. In addition to establishing the interindividual degree of variation within a broad panel of immune signaling responses, the possible association of any observed variation with demographic variables including age and race was investigated. Using half of the donors as a training set, multiple age- and race-associated variations in signaling responses in discrete cell subsets were identified, and several were subsequently confirmed in the remaining samples (test set). Such associations may provide insight into age-related immune alterations associated with high infection rates and diminished protection following vaccination and into the basis for ethnic differences in autoimmune disease incidence and treatment response. SCNP allowed for the generation of a functional map of healthy immune cell signaling responses that can provide clinically relevant information regarding both the mechanisms underlying immune pathological conditions and the selection and effect of therapeutics.

Molecules with similar shapes and features often have similar biological activity. Several computational approaches search chemical databases for new leads or templates based on overall molecular shape similarity. However, active molecules often present critical subshapes that are required for binding, which may be missed by comparing overall shape similarity. We present a new approach to compare molecular shapes of different sizes and to calculate subshape similarity. We developed a skeletal representation of the shape which is topologically unrelated to covalent chemical connectivity. This simplifies rotational and translational sampling. We test initial possible alignments by matching similar triangles. This triangle-matching filter rapidly eliminates most geometrically impossible matches. Surviving matches are filtered further in successive stages. These stages involve direction, feature, and shape matching procedures. Our approach is applied to several situations demonstrating lead discovery and evolution.

Background Single Cell Network Profiling (SCNP) is used to simultaneously measure the effects of modulators on signaling networks at the single cell level. SCNP-based biomarker assays predictive of response to induction therapy and relapse risk in AML patients are being developed. Such assays have typically utilized bonemarrow (BM) as the sample source of blasts. Since circulating peripheral blasts are detectable in ~65% of AML patients and peripheral blood (PB) sampling is less invasive than BM sampling, this study was performed to assess the effect of sample source on AML blasts signaling as measured in SCNP assay. Methods SCNP using multiparametric flow cytometry was used to evaluate the activation state of intracellular signaling molecules in leukemic blasts under basal conditions and after treatment with modulators in 46 pairs of BM mononuclear cells/PB mononuclear cells. The relationship between readouts of modulated intracellular proteins (“nodes”) was measured using linear regression, Bland-Altman method and Lin’s concordance correlation coefficient. Results The majority (156/161) of signaling nodes show strong correlations between paired PB and BM samples independently from the statistical method used. Notable exceptions were two PB samples with almost undetectable levels of circulating blasts compared to paired BM samples. Conclusions Our results demonstrate that specimen source (BM or PB) does not significantly affect proteomic signaling in patients with AML and circulating blasts. The ability to use PB as a sample source will facilitate the monitoring of cellular signaling effects following administration of targeted therapies and at time-points when BM aspirates are not clinically justifiable.