Executive forum: Vital statistics

Greater standardization and more advanced collaboration and data analyzing tools are helping propel scientists towards that goal. So what challenges still lie ahead? NGP asked Dave Champagne, Vice President and General Manager of Informatics at Thermo Electron Corportion; Bill Harten, founder of UNIConnect, and Pat Martell, Director of Waters Laboratory Informatics, and Larry Norder, Vice President and General Manager of Jubilant Discovery Services.

There is obviously a need for the pharma industry to leverage knowledge to improve the efficiency of the drug discovery cycle. How is this being achieved through the exploration and analysis of large quantities of data to discover patterns?

PM. The amount of data being generated continues to grow, especially as new systems become more available to shorten sample times and provide greater sensitivity and resolution. The real challenge in drug discovery is not getting the data but trying to analyze it and then, when you go into trials, how do you get all of the information so that you can start to visualize trends and analysis. A lot has to do with understanding the information you’ve captured and, once analyzed, making informed decisions based upon that.

Another challenge is understanding the details of what is going on in a cell at a specific moment that has been treated or affected by a compound. The more we can tell about that particular interaction at that point, the better scientists can determine the effectiveness of their efforts and experimentation. The quickest way scientists can understand interactions in the cell is for them to understand and optimize the information processing that is associated with the analysis. What we’re seeing is the melding of information management or processing and science to the highest extent that we’ve ever seen in the industry today to make the next breakthrough.

BH. Machine learning algorithms use a set of training data with known classifications to discover which patterns are significant. These algorithms are cracking many kinds of problems we couldn’t deal with before. For one of our clients we recently pitted a respected expert against a battery of 60 different machine learning algorithms to discover which was most powerful in a classification problem. Astonishingly, the expert placed only 12th. The best machine-learning algorithms performed much better, making only half as many classification errors, and of course they did it much faster and more consistently. And they didn’t complain when we asked them to do it all over again with new rules and new data.

More machine-learning algorithms are coming out all the time. We have found it’s very powerful to harness them into an automated comparison framework that pits them against each other to determine which ones work best for a given problem. The metrics include error measurements for false positives and negatives, ability to explain the classification criteria, ability to recognize when a pattern cannot be reliably classified, and compute time, which is critical for analyzing large, complex datasets.

LN. As the cost of generating data and information has decreased, the amount of data available to research teams has risen astronomically. It has been likened to trying to drink from a firehose. Large sets of previously thought unrelated data have been found to provide valuable insights. The challenges are in assembling all the data in a rational way and asking the right questions. Synthesizing these large databases requires insight to the mechanisms by which they will be mined, as well as insight into the chemical and biological significance of the data itself. We can broadly characterize the categories into two groups – bioinformatics and chemoinformatics.

DC. There are a number of companies and researchers working on using high power computing and pattern recognition algorithms to improve the very early stages of drug discovery, but we do not play in that space right now. At Thermo, we are exposing knowledge for greater efficiency by moving to open standards to harmonize processes, create more consistent reporting and provide real-time availability.

In what ways is information currently tracked and how is this trend changing in terms of replacing customized programs?

BH. The problem is that the process steps and data are different for every lab, without exception. Each is as unique as a fingerprint. One-size-fits-all programs do both too much and too little, and still don’t implement the unique process – canned software that fits is a myth. It’s not enough to record the sample at the beginning and the results at the end. It’s the steps in between where resources are tracked and errors prevented rather than corrected afterward, but this requires monitoring every step as you go, requiring a custom solution. Some build their own, some pay to customize the program, but software customization isn’t ‘soft’, it’s hard, costly, and painful. But there is a better way.

Over the past decade we have discovered universal patterns in the way labs think about processes and data, including patterns for defining steps, content flows, workflows, and the resources, reagents, and information to be validated and recorded at each step. We defined keywords and parameters that embody these patterns, programmed them into our UNIFlow engine, and enable the lab’s process specialist to invoke these patterns in a simple outline to configure a custom process using their own terminology, without custom programming. The result is a commercial off-the-shelf system with a detailed user-modifiable configuration. Customizing this way is fast, reliable, and allows a proper fit.

DC. Laboratory Information Management Systems (LIMS) are used to store, monitor, track and audit data, as well as other functions. The pharma industry is moving away from customizing LIMS – at great risk and expense to accommodate its needs – in favor of using commercial-off-the-shelf (COTS) solutions that deliver a much higher percentage of required functionality out of the box. At Thermo we offer four distinct LIMS to meet the needs across the entire pharmaceutical value chain from early discovery to production.One thing that is clear is that pharma companies are increasingly concerned with the tracking of research data for intellectual property (IP) protection, as well as for more extended information sharing. The need to protect IP is obvious; the need for improved information sharing is driven by a desire to improve R&D productivity and avoid redundant experimentation.

PM. We see chromatography data software solutions, LIMS packages, statistical packages, etc. but there are a lot of separate software solutions that scientists need to be able to use – not to mention that they have different systems that they need to be able to run. So what they have is an amalgamation of a lot of different programs they have to be conversant with and different user experiences with regard to instrumentation – and that doesn’t even touch upon the papers that they have to sift through and generate. When you look at that dynamic, customers are looking to realize product gains in how they can automate different processes. We recently announced a product for Waters Empower 2 Software called Method Validation Manager, which takes a very laborious paper-based task and the SOP that covers the process and automates it. Not only does that dramatically cut the time it takes to set up experiments but it also minimizes transcription errors, takes out all the paper-based processing and calculations, and puts the information into a compliant-ready environment so they can show traceability for the results – they know who did it, when they did it, what instruments they used, etc. in one application environment.

LN. The data is currently scattered all over – in public domain (hundreds of sites), patents (some time non-readable text), private houses, mostly in hardcopy or in some cases soft data yet to be integrated for future use. The trend is to organize the public data into easily queriable reliable data source, which can be integrated into internal data by appropriate software technology.

Eliminating errors and controlling processes is very important in tracking. How is this being achieved?

DC. The most common source of errors is manual data transcription; increased direct instrument interfacing has been a focus of laboratories for many years now, but due to the lack of standards between the many instrument vendors it has been difficult to implement at times. Thermo has been at the forefront of providing software to simplify and automate the interface between our LIMS, chromatography data systems (CDS) and laboratory instruments, as well as enterprise resource planning systems.

Errors also happen when data processing occurs outside of a controlled system. Errors in the calculations performed either on paper or in non-validated spreadsheets are difficult to detect, and late detection can often invalidate all the conclusions, requiring the relevant experiments to be repeated. Solutions like Thermo’s Galileo LIMS automate calculations to eliminate manual errors. Process control and elimination of these errors depends on implementing systems that provide more in-depth domain understanding and functionality. Pharma labs are now looking not only for commercial products, but also for commercial products which provide much more extensive domain functionality. When such solutions are selected as the standard for the global pharma enterprise, it harmonizes processes and leads to greater consistencies as well as broader dissemination of information. We are also seeing the need for compliance and auditing moving upstream in the development process as companies are more and more frequently using early research data in discussions with regulatory agencies and in submissions. The pharma industry benefits from vendors who can integrate data generated across the pharmaceutical discovery and development spectrum.

BH. Every worker, instrument, reagent and document goes through its own sub process that is just as critical as the main process but is too often ignored. Reagents have to be checked, workers require training and re-training, and instruments require periodic calibration and maintenance. It has become essential, even mandatory to have a system that ensures these processes have happened and to block a step if something isn’t right. This means entering and validating all the resources just before the step is actually performed. Our UNIFlow-based labs put barcodes on everything and track the lifecycles of the resources as well as the samples. Scanners and rapid input screens make this efficient, and workers feel a huge relief not having to manage so much in their heads and notebooks.

The challenge is that no two labs are the same; the steps and resources to track are always different, and defining these needs to be quick, reliable, and easy to change. Traditional software development cannot keep up, but having a simple language where lab people can manage this themselves makes it practical and affordable.

PM. The problem is that one person could interpret method validation SOP in one way and another person in another way, so the precision of the information may be close, it may not be – and that’s just trying to validate a method. The number of times that occurs in the development process in some companies could be up to 10 or 12. With the Method Validation Manager application, you can minimize those time savings by 40-85 percent. We’re helping customers save costs and optimize their investment in time and energy, and add critical value to that process by being able to automate the acquisition of the data, the processing of it and the traceability, all within a compliant environment.

LN. The errors in data obtained from the original source can be corrected by having good quality processes. However, this cannot rule out the occurrence of data errors in the public domain. This can be addressed if several other sets of published data can be analyzed (if any) related to the topic of interest. This can be achieved through good curation practice and database creation, intelligent data mining and analytical tools.

How can information be pulled together from different areas to assist the drug discovery process and lab capabilities?

BH. A study at MIT two decades ago showed that it is better to tolerate different models of information and to translate between them than to coerce an enterprise or an industry to use a single common model. We’ve applied this in our databases by providing generic tables for LIMS tracking and allowing these to be joined with lab-specific detail tables that represent the unique dimensions of the core problem. The key to marrying these diverse types of information is enterprise-wide discipline in assigning identification numbers to everything and using the same ID in each of the tables that pertain to the same entity. This approach allows information models to evolve independently while providing points of common ground to connect information from one to the other.

PM. I’ve talked to customers who may have as few as six analytical applications and others who may have over 20 applications. The information from each of those applications is somehow important to the effort at hand. The critical thing from a solution point of view is how can we help them aggregate that information and not only collect it into a repository, but ensure they are easily able to find that information when they need it and be able to use it in a more meaningful way. For example, a company may have chromatography data, mass spectrometry information and NMR data. Each one by itself is important but they might need to pull it all together in an INDA form that needs to be submitted for a drug application. With Waters SDMS (Scientific Data Management System), customers can pull those pieces of information automatically off the software and capture the meta data so they know what type of information it is, when it was collected and what method was used. At a later date, they are also able to search that information for specific compounds.

Another tool we have that adds another level of value is the eLab Notebook, which allows scientists to individually or collectively capture information from laboratory processes. In a paper-based laboratory, this information may not be captured by the corporation or available outside the immediate group.

DC. The gradual penetration of improved technology standards such as web services and XML is making data aggregation and transfer easier. However, because of the slow adoption of new software tools and technologies by vendors and laboratories, there will still be a great deal of manual data aggregation required to gain holistic understanding of drug development results. It is clear that the automation of this data aggregation, and presentation of it in a meaningful user interface, is a very high priority within the pharma industry. Thermo has adopted web services and service oriented architecture in our LIMS development in order to address these needs within the pharmaceutical industry.

LN. From target identification to clinical – data needs to be made accessible to researchers in a common backbone, both in modular fashion for appropriate stages and an integrated engine for use by research managers. One specific example that addresses one of the most pressing issues facing pharma – failure of compounds in the clinic – is integrating development stage data into the early stage design on compounds. In this way, the scientist can leverage what is known about pharmacodynamics and pharmacokinetics to ‘navigate’ away from liabilities that can lead to adverse events and failures in the clinic.

To further address the issue, pharma companies are turning to diagnostics coupled to the drug to identify patient populations where the drug is efficacious due to their unique metabolic and genetic profile. The success of Genentech’s Herceptin in the treatment of a subclass of breast cancer patients with a specific genetic mutation, validates this approach.

How is the industry minimizing communication gaps between scientists, researchers and IT solutions to reach targets?

BH. The great advances in information technology came as high-level languages harnessed the computers to plow through our challenging problems. Examples include SQL, the language of the database, Mathematica, HTML for the web, spreadsheets, and of course the many programming languages. Scientists and researchers are inventing their own new languages all the time to grapple with new problem domains; indeed, language is almost the measure of scientific progress. It’s empowering when one of these is conceived and embodied in software such that a ‘word is worth a thousand pictures’. This lets the computer do the repetitive and tedious part while liberating the human mind to imagine, invent and discover meaning.

The IT industry is buzzing right now about this domain-specific language (DSL) concept and powerful new tools to make DSLs easier to create. Our company’s UNIFlow engine is based on a DSL that lets a non-programmer specify detailed lab process requirements in just the right words and parameters. We’ve invented some of the tools to help create DSLs, and we expect to see more DSLs tackle the problems of our industry.

DC. What we’re seeing is greater standardization. When it comes to LIMS, pharmaceutical organizations often have a variety of systems. This stems from mergers and acquisitions, but also because individual lab managers often made the decision of which LIMS to purchase based on individual lab needs. Standardizing delivers the business benefits that help pharma reduce total cost of ownership, and if you select a COTS solution, the users can get the features and functionality they want at the lowest risk because of reduced customization to the system.

PM. When you look at the communication capabilities built into solutions like the eLab Notebook, it gives people access to collaboration possibilities with other scientists that have conducted research in your areas, outside your department – even across the globe.

LN. The industry is moving ahead to create data accession, data integration, data applications and solutions to reach scientists. An example of this trend is the proliferation of electronic notebook initiatives and products in the industry and, even more cogently, the development of public standards such as SBML, the structural biology mark-up language for organization of structural biology information into standardized searchable databases.

Overall, do you see advances in IT propelling informatics over the next few years?

PM. The role that IT is playing is becoming more important. If you think back 20 years ago, a lot of the computer solutions that were available were stand-alone workstation systems and IT didn’t really play much of a role in the lab. Now, the IT department is increasingly working with C-level executives to minimize the infrastructure. The challenge is to optimize the work process, the throughput and productivity.

When IT is engaged, it is there to standardize the application suite available to scientists, minimize and standardize the deployment, validation and training that will be necessary for larger organizations to have in regard to the applications they run; and they are looking to globalize so their other companies can access the information in a uniform way.

DC. Informatics is and will continue to be the process “glue” that enables workflows across the pharma enterprise while ensuring a high degree of automation and compliance.

LN. IT will serve its main purpose, giving a better shape to the integration of innumerable data (internal, external) with the appropriate backbone and application layers for end-users for timely use of the needed knowledge and data.