Research Grants/Fellowships/SBIR

Final Report: BESS, A System For Predicting The Biodegradability Of New Compounds

EPA Grant Number: R826114
Title: BESS, A System For Predicting The Biodegradability Of New Compounds
Investigators: Punch, William F.
Institution: Michigan State University
EPA Project Officer: Karn, Barbara
Project Period: November 1, 1997 through October 31, 2000
Project Amount: $285,063
RFA: Technology for a Sustainable Environment (1997) RFA Text |  Recipients Lists
Research Category: Sustainability , Pollution Prevention/Sustainable Development



The objective of this research project was to further develop a software system called BESS (Biodegradation Evaluation and Simulation System), which can predict the biodegradability of a compound based on the structural features of that compound and the prevailing environmental conditions. The approach pursued in the development of BESS is based on the iterative use of plausible enzymatic transformations that are hierarchically organized based on knowledge of microbial physiology and ecology. This organization reduces the potentially large number of enzymatic transformations that could apply to a compound, making the approach computationally feasible. Further, only those enzymatic transformations that are most likely to provide anabolic intermediates or energy to microorganisms and thus have evolved through processes of natural selection are used. This further reduces the complexity.

BESS employs a fundamentally different approach to the prediction of whether a given chemical will undergo biodegradation in the environment. The approach is based on the simulated application of plausible enzymatic transformations in an iterative manner. The enzymatic transformations are hierarchically organized, reflecting knowledge about the likelihood of any enzyme being applied. Such an organization reduces the potentially large number of enzymatic transformations that could apply, concentrating instead on those rules that are most likely to provide anabolic intermediates or energy (or some other benefit to microorganisms) and thus have evolved through processes of natural selection. Thus far, rules have been codified for a limited number of chemical classes, and prototype software has been developed that is capable of correctly simulating the biodegradation of chemicals for which rules have been entered. A user interface has been developed that permits the user to review the proposed path of degradation step-by-step, modify or enter rules to explore "what if" scenarios, and identify metabolites that may accumulate in the environment. In this research project, the goals are to: (1) expand the number and kinds of enzymatic transformations that are codified; (2) develop and test a rule hierarchy; and (3) use genetic algorithms as a tool to "discover" rules that suggest as-of-yet unencoded, or even undiscovered pathways, giving the system a great deal of robustness in dealing with novel compounds.

The accomplishments of the proposed research were to be:

  1. The BESS software package, which can predict whether a specific compound is likely to undergo biodegradation in the environment. BESS would serve at least three functions:
    • The use of BESS would allow biodegradability to be considered early in the product development process.
    • BESS could be used by regulatory agencies to guide decisions regarding proposed regulation and testing.
    • BESS could be used as a tutorial to teach users what features of chemicals or the environment restrict or promote biodegradation.
  2. A compilation of plausible biodegradation rules that, independent of BESS, will constitute an important resource for environmental scientists.
  3. A "learning system" that could be applied to new cases, where degradation results were known but plausible pathways were not. BESS would be able to predict plausible pathways based on known pathways and structures.

Summary/Accomplishments (Outputs/Outcomes):

The BESS Software

Two rounds of construction were completed for the software. The first was a prototype version, written in Smalltalk, and the second a more refined version written in C++ to increase the efficiency of the prediction and to make porting the package easier.

Porting BESS to Procter & Gamble

The major accomplishment for this last cycle was a porting of the entire project to the Procter & Gamble site at Miami Valley and bringing up the system there for evaluation. This involved a number of steps, including:

  • Porting the database (which was in ISISbase) to a more usable database format. Work is ongoing to port the entire database (about 1,100 rules) to Oracle for use by the system.

  • Bring up the BESS program on a Procter & Gamble Web server. The portability of the codebase was tested by moving it to a Linux environment for testing with Procter & Gamble.

  • Creating visualization of the created structures. Procter & Gamble had access to a number of toolkits that allowed for display of arbitrary SMILES strings (which is what BESS works with presently) into 2-D models on the Web page for better visualization. This involved a lot of work but made the usability significantly better for local chemists.

  • A non client-server version of BESS was created in Visual Basic so that both users and developers could work with a non-networked system for experimentation.

  • Administrative issues regarding updates to rulesets are being examined for the expansion of the ruleset inside of Procter & Gamble.

Procter & Gamble is presently evaluating BESS for both internal use and possible fielding on their public Web site as a public service to the community.

The BESS Rule Base

Addition of the rule base of BESS has continued over the course of the project, though this has proved more difficult than anticipated. During the first 2 years, about 200 rules garnered by research assistants from the literature were entered. This process continued over the remainder of the grant. The Marty Alexander database (approximately 2,000 rules) was added, making about 3,000 or so rules available in BESS as of the last cycle. Now that the system is being field tested by Procter & Gamble using a local rule editor, it remains to be seen how many local rules will be added to the system.

The "Learning" System

A description of the learning problem that was addressed follows. If external testing indicated that a chemical would degrade, but the present BESS system could not find a plausible degradation path, the gap would be filled in. This was done by allowing the present BESS to degrade the chemical as far as possible. The intermediate that could not be further degraded was tested against "plausible rules," generated from the present rule set, to see if the addition of a plausible rule would indeed allow the degradation to run to completion, as indicated by the external testing. The minimal plausible rule, or set of such rules, that would allow degradation could then be examined by experts to see if they were truly plausible, potentially positing a new degradation reaction.

A prototype learning system was created based on a genetic algorithm approach to the problem. New rules could be generated, but the simulations proved very slow on the early Smalltalk implementation, making it difficult to evaluate. Porting the prototype to the final C++ version was not successful because the focus was on accomplishing the first two goals. It remains an intriguing possibility that should be pursued.

Journal Articles:

No journal articles submitted with this report: View all 3 publications for this project

Supplemental Keywords:

biodegradation, simulation, computer science, engineering, environmentally conscious manufacturing., RFA, Scientific Discipline, Sustainable Industry/Business, Environmental Chemistry, Sustainable Environment, Technology for Sustainable Environment, Biochemistry, computational simulations, biodegradation evaluation system, decision making, software system, computer science, microorganisms, modeling, enzyme transformations, biotechnology, computer generated alternatives

Relevant Websites: Exit EPA icon

Progress and Final Reports:
Original Abstract
1999 Progress Report
2000 Progress Report