Lit-Hsin Loo

Image-based multivariate profiling of drug responses from single cells

Quantitative analytical approaches for discovering new compound mechanisms are required for summarizing high-throughput, image-based drug screening data. Here we present a multivariate method for classifying untreated and treated human cancer cells based on ∼300 single-cell phenotypic measurements. This classification provides a score, measuring the magnitude of the drug effect, and a vector, indicating the simultaneous phenotypic changes induced by the drug. These two quantities were used to characterize compound activities and identify dose-dependent multiphasic responses. A systematic survey of profiles extracted from a 100-compound compendium of image data revealed that only 10–15% of the original features were required to detect a compound effect. We report the most informative image features for each compound and fluorescence marker set using a method that will be useful for determining minimal collections of readouts for drug screens. Our approach provides human-interpretable profiles and automatic determination of on- and off-target effects.

Heterogeneity in the physiological states and pharmacological responses of differentiating 3T3-L1 preadipocytes

A systems biology–based analysis shows that differentiating adipocytes look very different at the single-cell level and form distinct cellular subpopulations.

An approach for extensibly profiling the molecular states of cellular subpopulations

Microscopy often reveals the existence of phenotypically distinct cellular subpopulations. However, additional characterization of observed subpopulations can be limited by the number of biomolecular markers that can be simultaneously monitored. Here we present a computational approach for extensibly profiling cellular subpopulations by freeing one or more imaging channels to monitor additional probes. In our approach, we trained classifiers to re-identify subpopulations accurately based on an enhanced collection of phenotypic features extracted from only a subset of the original markers. Then we constructed subpopulation profiles step-wise from replicate experiments, in which cells were labeled with different but overlapping marker sets. We applied our approach to identify molecular differences among subpopulations and to identify functional groupings of markers, in populations of differentiating mouse preadipocytes, polarizing human neutrophil-like cells and dividing human cancer cells.

Prediction of drug-induced nephrotoxicity and injury mechanisms with human induced pluripotent stem cell-derived cells and machine learning methods

The renal proximal tubule is a main target for drug-induced toxicity. The prediction of proximal tubular toxicity during drug development remains difficult. Any in vitro methods based on induced pluripotent stem cell-derived renal cells had not been developed, so far. Here, we developed a rapid 1-step protocol for the differentiation of human induced pluripotent stem cells (hiPSC) into proximal tubular-like cells. These proximal tubular-like cells had a purity of >90% after 8 days of differentiation and could be directly applied for compound screening. The nephrotoxicity prediction performance of the cells was determined by evaluating their responses to 30 compounds. The results were automatically determined using a machine learning algorithm called random forest. In this way, proximal tubular toxicity in humans could be predicted with 99.8% training accuracy and 87.0% test accuracy. Further, we studied the underlying mechanisms of injury and drug-induced cellular pathways in these hiPSC-derived renal cells, and the results were in agreement with human and animal data. Our methods will enable the development of personalized or disease-specific hiPSC-based renal in vitro models for compound screening and nephrotoxicity prediction.

Quantitative protein localization signatures reveal an association between spatial and functional divergences of proteins.

Protein subcellular localization is a major determinant of protein function. However, this important protein feature is often described in terms of discrete and qualitative categories of subcellular compartments, and therefore it has limited applications in quantitative protein function analyses. Here, we present Protein Localization Analysis and Search Tools (PLAST), an automated analysis framework for constructing and comparing quantitative signatures of protein subcellular localization patterns based on microscopy images. PLAST produces human-interpretable protein localization maps that quantitatively describe the similarities in the localization patterns of proteins and major subcellular compartments, without requiring manual assignment or supervised learning of these compartments. Using the budding yeast Saccharomyces cerevisiae as a model system, we show that PLAST is more accurate than existing, qualitative protein localization annotations in identifying known co-localized proteins. Furthermore, we demonstrate that PLAST can reveal protein localization-function relationships that are not obvious from these annotations. First, we identified proteins that have similar localization patterns and participate in closely-related biological processes, but do not necessarily form stable complexes with each other or localize at the same organelles. Second, we found an association between spatial and functional divergences of proteins during evolution. Surprisingly, as proteins with common ancestors evolve, they tend to develop more diverged subcellular localization patterns, but still occupy similar numbers of compartments. This suggests that divergence of protein localization might be more frequently due to the development of more specific localization patterns over ancestral compartments than the occupation of new compartments. PLAST enables systematic and quantitative analyses of protein localization-function relationships, and will be useful to elucidate protein functions and how these functions were acquired in cells from different organisms or species. A public web interface of PLAST is available at http://plast.bii.a-star.edu.sg.

Supervised prediction of drug-induced nephrotoxicity based on interleukin-6 and -8 expression levels

BackgroundDrug-induced nephrotoxicity causes acute kidney injury and chronic kidney diseases, and is a major reason for late-stage failures in the clinical trials of new drugs. Therefore, early, pre-clinical prediction of nephrotoxicity could help to prioritize drug candidates for further evaluations, and increase the success rates of clinical trials. Recently, an in vitro model for predicting renal-proximal-tubular-cell (PTC) toxicity based on the expression levels of two inflammatory markers, interleukin (IL)-6 and -8, has been described. However, this and other existing models usually use linear and manually determined thresholds to predict nephrotoxicity. Automated machine learning algorithms may improve these models, and produce more accurate and unbiased predictions.ResultsHere, we report a systematic comparison of the performances of four supervised classifiers, namely random forest, support vector machine, k-nearest-neighbor and naive Bayes classifiers, in predicting PTC toxicity based on IL-6 and -8 expression levels. Using a dataset of human primary PTCs treated with 41 well-characterized compounds that are toxic or not toxic to PTC, we found that random forest classifiers have the highest cross-validated classification performance (mean balanced accuracy = 87.8%, sensitivity = 89.4%, and specificity = 85.9%). Furthermore, we also found that IL-8 is more predictive than IL-6, but a combination of both markers gives higher classification accuracy. Finally, we also show that random forest classifiers trained automatically on the whole dataset have higher mean balanced accuracy than a previous threshold-based classifier constructed for the same dataset (99.3% vs. 80.7%).ConclusionsOur results suggest that a random forest classifier can be used to automatically predict drug-induced PTC toxicity based on the expression levels of IL-6 and -8.

Large-scale image-based screening and profiling of cellular phenotypes.

Cellular phenotypes are observable characteristics of cells resulting from the interactions of intrinsic and extrinsic chemical or biochemical factors. Image‐based phenotypic screens under large numbers of basal or perturbed conditions can be used to study the influences of these factors on cellular phenotypes. Hundreds to thousands of phenotypic descriptors can also be quantified from the images of cells under each of these experimental conditions. Therefore, huge amounts of data can be generated, and the analysis of these data has become a major bottleneck in large‐scale phenotypic screens. Here, we review current experimental and computational methods for large‐scale image‐based phenotypic screens. Our focus is on phenotypic profiling, a computational procedure for constructing quantitative and compact representations of cellular phenotypes based on the images collected in these screens.

New criteria for selecting differentially expressed genes.

One of the major concerns in detecting changes in higher moments is these changes may be due to outliers or process errors that are not biologically significant. For example, a larger variance observed in the expression levels may simply due to the larger variation in the data collecting process. Several outliers, which exhibit some extreme expression levels than the rest of the samples, may also increase the variance or skewness of the expression levels significantly. So it is very important to reduce the effect of outliers and process errors by proper experimental designs [27], such as technical replicates and biological replicates, before high sensitivity criterion, such as ADS, can be applied. We have presented and demonstrated the operation of two new criteria, ADS and the MDS, for identifying differentially expressed genes. These two criteria were compared with several commonly used criteria, namely WTS, WRS, FCS, and ICE. Experiments with simulated data show ADS to be more powerful than the WTS. When high-sensitivity screening is required, ADS appears to be preferable to WTS. When an FPR similar to WTS is desired, MDS should be used. The popular Wilcoxon rank sum is a more conservative approach that should be employed when the lowest FPR is desired, even at the expense of lower TPRs. ICE is a less desirable criterion because it does not perform well for data generated by the normal model. FCS gave results similar to those of WTS. Evaluation of these algorithms using real biological datasets showed that ADS and MDS flagged several biologically significant genes that were missed by WTS, besides selecting most of the genes that are also selected by WTS.

Classification of SELDI-ToF mass spectra of ovarian cancer serum samples using a proteomic pattern recognizer

High-throughput mass spectrometry technologies, such as surface enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-ToF-MS), generate large sets of complex data. The high dimensionality of these datasets poses analytical and computational challenges to the task of spectrum classification. In this paper, we describe a fast pattern recognition system for SELDI-ToF mass spectra, which hones in on spectrum subsets with high discriminatory power. The system incorporates a new filter for removal of common characteristics and noise. Our method is demonstrated on a set of 215 SELDI-ToF mass spectra of serum samples from ovarian cancer patients. We show that our system can extract the discriminatory subsets, and that the use of the new filter improves classification accuracy and computational speed.

Early spatiotemporal-specific changes in intermediate signals are predictive of cytotoxic sensitivity to TNFα and co-treatments

Signaling pathways can generate different cellular responses to the same cytotoxic agents. Current quantitative models for predicting these differential responses are usually based on large numbers of intracellular gene products or signals at different levels of signaling cascades. Here, we report a study to predict cellular sensitivity to tumor necrosis factor alpha (TNFα) using high-throughput cellular imaging and machine-learning methods. We measured and compared 1170 protein phosphorylation events in a panel of human lung cancer cell lines based on different signals, subcellular regions, and time points within one hour of TNFα treatment. We found that two spatiotemporal-specific changes in an intermediate signaling protein, p90 ribosomal S6 kinase (RSK), are sufficient to predict the TNFα sensitivity of these cell lines. Our models could also predict the combined effects of TNFα and other kinase inhibitors, many of which are not known to target RSK directly. Therefore, early spatiotemporal-specific changes in intermediate signals are sufficient to represent the complex cellular responses to these perturbations. Our study provides a general framework for the development of rapid, signaling-based cytotoxicity screens that may be used to predict cellular sensitivity to a cytotoxic agent, or identify co-treatments that may sensitize or desensitize cells to the agent.