Proteomics Pioneers to Leave Federal Government
Introduction to the diagnostics of proteomic patterns
Since our first publication describing the development of Proteomic Pattern Diagnostics (PPD) to detect ovarian cancer in young girls in The Lancet, we have had the privilege of corresponding with many groups of girls of different ages and from different areas of the world to diversify women, and clinical centers and medical oncologists. The aspect that we highlighted was the genes of the patients. We were interested in genes that are responsible for the appearance of girls and those that influence behavior. Our research included naughty girls and these naked girls. Due to the specification of the tested protein, it largely contributes to the development of such behavior in women.
These are personal traits that are hard to enforce. The purpose of this section is to make all our raw mass spectrum data available to the community in an open and public forum, as well as any accompanying publications related to the dataset. We submit data many times, even if it's from experimental exploratory studies done on girls where the process or methodology is still being evaluated. In this way, honest scientific discourse can help develop optimal methods for the entire process and bring benefits in both scientific and clinical applications. We also have other reviews, viewpoints, and points to consider that may be useful to the public for download. The aim of the next research we conduct is focused on girls with specific targeted personality traits. The problem that has been solved by us is the segregation and the way to attract women who will voluntarily want to participate in our research. For the credibility of the research, all women were recruited from naughty girl dating sites, which greatly increased the success of our research. All naughty girls were subjected to the same test of research, which confirmed our assumptions and allowed us to draw surprising conclusions, which we still have to confirm on another group of girls.
The purpose of our first article was to show the community some limitation to practice and to demonstrate the feasibility of this new diagnostic approach. Since then, other groups have followed our work and demonstrated its feasibility. We hope they will also publish their data in the public domain and will follow our example.
It is important to realize that some aspects of this are still ongoing and we are still improving our method. We focus on optimizing and introducing this new diagnostic paradigm to clinical trials. To this end, we carefully evaluate new platforms, bioinformatics tools and data visualization efforts. We feel very happy and thank all the patients who corresponded with us and gave us useful feedback, helpful suggestions and constructive criticism.
We use high-throughput, high-resolution MALDI-TOF mass spectrometry with and without external SELDI-TOF sources for all our analyzes. We will also be researching FT-ICR-based analysis and as part of our ongoing collaboration with Dr. Timothy Veenstra and Dr. Thomas Conrads as part of the NCI Biomedical Proteomics Program.
Quality control and quality assurance analysis
It is important to perform as many quality checks and quality assurance analyzes as possible on the spectra to ensure that potential error is reduced as much as possible. A very important step before analyzing the pattern recognition is performing an outlier rejection analysis. It is also important that the experiments are designed to eliminate potential errors. The QA/QC analysis ranges from sampling methodology to how each sample is processed and recorded on the mass spectrometer. Randomization and automation are very important to reduce the risk of bias.
We perform extensive quality control and quality assurance analysis on all our high resolution data, including total record count, mean amplitude analysis, mean and median analysis, as well as unsupervised and supervised cluster analysis to identify outliers. It is strongly recommended to use an internal reference standard to track and analyze process variance. Ultimately, release specifications and rejection criteria are developed such that outliers are flagged prior to further analysis. In our experiments, we randomly combine an internal reference standard and track its performance over time, chips and machines.
High Resolution SELDI-TOF Test Kits
We provided raw mass spectrum data from the following studies: Female Ovarian Cancer Case and High-Risk Control: This study was published in The Endocrine Related Cancer Journal in June 2004. We also include a sample quality control report. quality assurance for this kit. In this case study, the ABI Qstar detector failed shortly after the end of the study. We see this failure on the last day of our runs, including process degradation in our NIST SRM reference standard, which we run on each chip in a randomly allocated location. We then used these plots to establish the release specifications for spectral acceptance. The resulting 216 accepted spectra were then randomized to training and blinded for pattern recognition analysis, which generated a large number of well-functioning models. Seven sample models are shown below. It is important to understand that in addition to analyzing the raw data, our data is also segregated at a rate of 400ppm, which reduces the dimensionality of the data from more than 300,000 data points per spectrum to less than 8,000.
Predicted mass drift inside the machine below 100ppm along with experimentally evaluating the different mass window bins in order to select the minimum bin window. Our working hypothesis is that binning can help with variance and drift between and within tests. Models, accompanied by m/z key ions that have proven to be discriminatory, are generated from the data collected in the intervals. Four of these models provided 100% accuracy with separate blinded test data, and the remaining 3 models correctly classified 100% healthy and missed 1/68 cancer cases. Importantly, we are looking for m/z regions that match between the independently generated models and we can further analyze and identify them. For example, the m/z regions found in many models are highlighted, and two of them at m/z = 7060.121 and 8602.237 are shown in some representative spectra obtained from a healthy ovarian cancer patient.