Universal automated classification of the acoustic startle reflex using machine learning.

The startle reflex (SR), a robust, motor response elicited by an intense auditory, visual, or somatosensory stimulus has been widely used as a tool to assess psychophysiology in humans and animals for almost a century in diverse fields such as schizophrenia, bipolar disorder, hearing loss, and tinnitus. Previously, SR waveforms have been ignored, or assessed with basic statistical techniques and/or simple template matching paradigms. This has led to considerable variability in SR studies from different laboratories, and species. In an effort to standardize SR assessment methods, we developed a machine learning algorithm and workflow to automatically classify SR waveforms in virtually any animal model including mice, rats, guinea pigs, and gerbils obtained with various paradigms and modalities from several laboratories. The universal features common to SR waveforms of various species and paradigms are examined and discussed in the context of each animal model. The procedure describes common results using the SR across species and how to fully implement the open-source R implementation. Since SR is widely used to investigate toxicological or pharmaceutical efficacy, a detailed and universal SR waveform classification protocol should be developed to aid in standardizing SR assessment procedures across different laboratories and species. This machine learning-based method will improve data reliability and translatability between labs that use the startle reflex paradigm.

Real-time impedance feedback to enhance cutaneous gene electrotransfer in a murine skin model.

Electric field mediated gene delivery methods have the ability to efficiently transfect cells in vivo with an excellent safety profile. The method has historically used a fixed number of electric pulses with identical characteristics in induce delivery. Electrical treatment does not typically compensate for subject-to-subject variation and other differences. This study was designed to investigate if delivery/expression could be increased using a novel electropulsation method that compensated for variation using real-time electrical impedance measurements. The method involved delivering plasmid DNA encoding luciferase to murine skin. Tissue impedance in a 1-3 KHz range was measured before electric pulses were applied. Impedance was also measured after each successive pulse. Pulsation was stopped when impedance values were reduced by either 80% or 95% relative to prepulse values. Standard/fixed pulsing parameters were also used for comparison. The results indicated that up to 15-fold increases in luciferase expression could be obtained when electrical treatment was ceased based upon impedance reductions. Furthermore, peak expression levels of all treatment groups pulsed using the novel pulsing method were statistically higher than those that employed standard pulsing. These results strongly suggest that applying pulses until a defined impedance-based endpoint results in higher expression.

Machine learning, waveform preprocessing and feature extraction methods for classification of acoustic startle waveforms.

The acoustic startle response (ASR) is an involuntary muscle reflex that occurs in response to a transient loud sound and is a highly-utilized method of assessing hearing status in animal models. Currently, a high level of variability exists in the recording and interpretation of ASRs due to the lack of standardization for collecting and analyzing these measures. An ensembled machine learning model was trained to predict whether an ASR waveform is a startle or non-startle using highly-predictive features extracted from normalized ASR waveforms collected from young adult CBA/CaJ mice. Features were extracted from the normalized waveform as well as the power spectral density estimates and continuous wavelet transforms of the normalized waveform. Machine learning models utilizing methods from different families of algorithms were individually trained and then ensembled together, resulting in an extremely robust model.•ASR waveforms were normalized using the mean and standard deviation computed before the startle elicitor was presented•9 machine learning algorithms from 4 different families of algorithms were individually trained using features extracted from the normalized ASR waveforms•Trained machine learning models were ensembled to produce an extremely robust classifier.

Automated classification of acoustic startle reflex waveforms in young CBA/CaJ mice using machine learning.

BACKGROUND: The acoustic startle response (ASR) is a simple reflex that results in a whole body motor response after animals hear a brief loud sound and is used as a multisensory tool across many disciplines. Unfortunately, a method of how to record, process, and analyze ASRs has yet to be standardized, leading to high variability in the collection, analysis, and interpretation of ASRs within and between laboratories. NEW METHOD: ASR waveforms collected from young adult CBA/CaJ mice were normalized with features extracted from the waveform, the resulting power spectral density estimates, and the continuous wavelet transforms. The features were then partitioned into training and test/validation sets. Machine learning methods from different families of algorithms were used to combine startle-related features into robust predictive models to predict whether an ASR waveform is a startle or non-startle. RESULTS: An ensemble of several machine learning models resulted in an extremely robust model to predict whether an ASR waveform is a startle or non-startle with a mean ROC of 0.9779, training accuracy of 0.9993, and testing accuracy of 0.9301. COMPARISON WITH EXISTING METHODS: ASR waveforms analyzed using the threshold and RMS techniques resulted in over 80% of accepted startles actually being non-startles when manually classified versus 2.2% for the machine learning method, resulting in statistically significant differences in ASR metrics (such as startle amplitude and pre-pulse inhibition) between classification methods. CONCLUSIONS: The machine learning approach presented in this paper can be adapted to nearly any ASR paradigm to accurately process, sort, and classify startle responses.

Germline cytoskeletal and extra-cellular matrix-related single nucleotide variations associated with distinct cancer survival rates.

BACKGROUND: Human mutagenesis has a large stochastic component. Thus, large coding regions, especially cytoskeletal and extra-cellular matrix protein (CECMP) coding regions are particularly vulnerable to mutations. Recent results have verified a high level of somatic mutations in the CECMP coding regions in the cancer genome atlas (TCGA), and a relatively common occurrence of germline, deleterious mutations in the TCGA breast cancer dataset. METHODS: The objective of this study was to determine the correlations of CECMP coding region, germline nucleotide variations with both overall survival (OS) and disease-free survival (DFS). TCGA, tumor and blood variant calling files (VCFs) were intersected to identify germline SNVs. SNVs were then annotated to determine potential consequences for amino acid (AA) residue biochemistry. RESULTS: Germline SNVs were matched against somatic tumor SNVs (i.e., tumor mutations) over twenty TCGA datasets to identify 23 germline-somatic matched, deleterious AA substitutions in coding regions for FLG, TTN, MUC4, and MUC17. CONCLUSIONS: The germline-somatic matched SNVs, in particular for MUC4, extensively implicated in cancer development, represented highly, statistically significant effects on OS and DFS survival rates. The above results contribute to the establishment of what is potentially a new class of inherited cancer-facilitating genes, namely dominant negative tumor suppressor proteins.

T-cell receptor-ß V and J usage, in combination with particular HLA class I and class II alleles, correlates with cancer survival patterns.

Class I and class II HLA proteins, respectively, have been associated with subsets of V(D)J usage resulting from recombination of the T-cell receptor (TCR) genes. Additionally, particular HLA alleles, in combination with dominant TCR V(D)J recombinations, have been associated with several autoimmune diseases. The recovery of TCR recombination reads from tumor specimen exome files has allowed rapid and extensive assessments of V(D)J usage, likely for cancer resident T-cells, across relatively large cancer datasets. The results from this approach, in this report, have permitted an extensive alignment of TCR-ß VDJ usage and HLA class I and II alleles. Results indicate the correlation of both better and worse cancer survival rates with particular TCR-ß, V and J usage-HLA allele combinations, with differences in median survival times ranging from 7 to 130 months, depending on the cancer and the specific TCR-ß V and J usage/HLA class allele combination.

Cytoskeleton and ECM tumor mutant peptides: Increased protease sensitivities and potential consequences for the HLA class I mutant epitope reservoir.

Cytoskeleton and extracellular matrix-related proteins (CECMPs) represent the most common class of cancer mutants, owing to the large size of their coding regions and the randomness of mutagenesis. We used a bioinformatics approach to assess the impact of amino acid (AA) substitutions on the sensitivity of CECMPs to proteases relevant to melanoma and on the binding affinities for HLA class I. CECMP peptides with AA substitutions overwhelmingly reflect increased sensitivity to proteases implicated in melanoma development (MME, CTSS, MMP2, CTSD, CTSL) in comparison to the wild-type peptide sequences. Furthermore, peptides with AA substitutions representing increased peptide protease sensitivity also represented relatively high binding affinities for HLA class I allelic variants. These analyses raise the question of whether increased protease sensitivity, of mutant cancer peptides represents a significant increase in the availability of cancer mutant, HLA class I epitopes and a hitherto unappreciated aspect of cancer cell immunogenicity, particularly in the case of melanoma?

Lung tumor exome files with T-cell receptor recombinations: a mouse model of T-cell infiltrates reflecting mutation burdens.

Tumor exomes and RNASeq data were originally intended for obtaining tumor mutations and gene expression profiles, respectively. However, recent work has determined that tumor exome and RNAseq read files contain reads representing T-cell and B-cell receptor (TcR and BcR) recombinations, presumably due to infiltrating lymphocytes. Furthermore, the recovery of immune receptor recombination reads has demonstrated correlations with specific, previously appreciated aspects of tumor immunology. To further understand the usefulness of recovering TcR and BcR recombinations from tumor exome files, we developed a scripted algorithm for recovery of reads representing these recombinations from a previously described mouse model of lung tumorigenesis. Results indicated that exomes representing lung adenomas reveal significantly more TcR recombinations than do exomes from lung adenocarcinomas; and that exome files representing high mutation adenomas, arising from chemical mutagens, have more TcR recombinations than do exome files from low mutation adenomas arising from an activating Kras mutation. The latter results were also consistent with a similar analysis performed on human lung adenocarcinoma exomes. The mouse and human results for obtaining TcR recombination reads from tumor specimen exomes are consistent with human tumor biology results indicating that adenomas and high mutation cancers are sites of high immune activity. The results indicate hitherto unappreciated opportunities for the use of tumor specimen exome files, particularly from experimental animal models, to study the connection between the adenoma stage of tumorigenesis, or high cancer mutation rates, and high level lymphocyte infiltrates.

Impedance spectroscopy as an indicator for successful in vivo electric field mediated gene delivery in a murine model.

In vivo gene electro transfer technology has been very successful both in animal models and in clinical trials over the past 20years. However, variable transfection efficiencies can produce inconsistent outcomes. This can be due to differences in tissue architecture and/or chemical composition which may effectively create unique biological environments from subject to subject that may respond differently to the identical electric pulses. This study investigates the integration of impedance spectroscopy into the gene electro transfer process to measure murine skin impedance spectra before, during (after pulse delivery), and after gene electro transfer pulse application to determine if changes in impedance correlate with reporter gene expression. Both post-treatment impedance spectra and gene expression were dependent upon the applied electric field strength. These results indicate that alterations in tissue impedance produced by the applied electric field represent an excellent parameter to predict degrees of transfection and gene expression. These results could ultimately be used to alter pulsing parameters in order to optimize delivery/expression.

A Novel Approach to Evaluating Cancer Driver Gene Mutation Densities: Cytoskeleton-related Gene Candidates.

BACKGROUND: Oncoprotein genes are over-represented in statically defined, low mutation-frequency fractions of cancer genome atlas (TCGA) datasets, consistent with a higher driver mutation density. MATERIALS AND METHODS: We developed a "continuously variable fraction" (CVF) approach to defining high and low mutation-frequency groups. RESULTS AND CONCLUSION: Using the CVF approach, an oncoprotein set was shown to be associated with a TCGA, low mutation-frequency group in nine distinct cancer types, versus six, for statically defined sets; and a tumor-suppressor set was over-represented in the low mutation-frequency group in seven cancer types, notably including BRCA. The CVF approach identified single-mutation driver candidates, such as BRAF V600E in the thyroid cancer dataset. The CVF approach allowed investigation of cytoskeletal protein-related coding regions (CPCRs), leading to the conclusion that mutation of CPCRs occurs at a statistically significant, higher density in low mutation-frequency groups. Supporting online material for this article can be found at www.universityseminarassociates.com/Supporting_online_material_for_scholarly_pubs.php.