Bioengineers facing conflicting data interpretations. How can you ensure accurate results in your research?

I’m coming from a microscopy imaging perspective. New tools need to be correlated with existing methods. Many of the points in this discussion also apply to this arena—and its cousin, medical (radiology, etc.) imaging.

When faced with conflicting data interpretations as a bioengineer, it's important to start by revisiting the data to verify its accuracy and ensure that all analysis methods are consistent. Encourage open discussion among team members to explore different interpretations and understand the basis for each perspective. Use additional experiments or replicate studies to confirm findings and resolve discrepancies. Document all interpretations and methodologies clearly to provide a comprehensive view of the data. Seek help from professionals outside the team if needed. In my experience, a collaborative approach, where all viewpoints are considered and tested, helps ensure the most accurate results and strengthens the research outcomes.

First ensure data quality, use multiple analytical methods, maintain transparency in records of analytical methods, collaborate to get peer-reviewed, and make sure the results are reproducible.

To ensure accurate data analysis in bioengineering research, it is essential to implement suitable controls, randomization, and replication to minimize bias and variability. Selecting the appropriate statistical tests based on the data type and research objectives is crucial, ensuring that all test assumptions are met. Thorough preprocessing of data to address missing values, outliers, and inconsistencies is also necessary. Keeping detailed records of methods and analyses facilitates reproducibility, while the use of independent datasets or alternative methods aids in validation. Presenting findings to peers for getting their perspective and employing validated software for data analysis ensures accuracy!

In my experience, conflicting interpretations are more common than not. If there were no conflicting interpretations, it would likely mean that the problem had already been solved, and we would be ready to move on to the next challenge. I tell my students that their primary responsibility is to draw conclusions that are consistent with the data. This becomes even more crucial — and difficult — when the results are unexpected. It is also essential to clearly state your conclusions, explain how the data support them, and acknowledge where the conclusions may be in conflict with each other. After all, science is an iterative process, and each cycle of discovery is driven by clear, objective insights gained from previous work.

A good scientist always has good ethics. A great part of being ethical is never publishing without ensuring your data is reproducible. A lot of advice these days I read on the Internet is where Pi's tell their students "never reproduce a perfect result". Don't listen to that advice. That is bad scientific training. Someone will reproduce your results and report it and make you retract your paper downstream if it is not reproducible. Also please do anything else with your life and don't be a scientist if you are going to be unethical.

Science = replication; that's a fundamental in the scientific method. However, bioengineering is engineering, not science. That means studies only need to be replicated to the extent that they validate the solution meets the specifications. Where I've seen the biggest challenges in my career is the hope that a few repeats are enough to replicate medical outcomes. This is rarely the case. The overwhelming majority of therapeutic and diagnostic devices I've worked on require many, many repeats over a wide variety of conditions to establish the confidence that is expected from medicine in the U.S. and from executives that operate the company.

Conduct experiments multiple times or review other studies that replicate your data to ensure the accuracy and reliability of your findings, eliminating the possibility of anomalies or errors.

Understand the Original Study: Before attempting to replicate a study, thoroughly review the original research, including the methods, materials, and analysis techniques used. This understanding is essential for accurate replication. Assess Feasibility: Determine if the original study can be replicated with the resources and equipment available to you. If specific materials or technologies are not accessible, consider modifying the approach while maintaining the core elements of the study.

Define Control Groups Clearly: Clearly delineate the types of control groups you will use, such as positive controls (where you expect a known outcome) and negative controls (where no effect is expected). This helps establish a solid baseline for comparison. Match Experimental Conditions: Ensure that control groups are subjected to the same experimental conditions as your test groups, including environmental factors, timing, and sample handling. This ensures that any observed differences are attributable to the variable being tested. Use Sufficient Sample Sizes: Employ an adequate number of samples in both experimental and control groups

In my career, all the projects I've worked on expect controls, but achieving them is another matter. The deep questions arise when engineering is defined by one set of observations/specifications and biotech/medicine have a different set of observations/specifications. Engineering is often characterized by a series of intermediate specifications (e.g. the tolerance on individual parts to achieve a system tolerance). However, biotech/medical applications don't always have access to intermediate controls. As such, they are often outcome based. Outcome based controls are very risky and expensive. Great thought is required to balance cost, risk, and certainty in a cost/time efficient manner.

You are joking right? Why is this even a question? How will you ever analyze your data and know that your experiment is correct without controls. Like the gentleman before me wrote, go back to school if you don't understand the value of controls.

Implement control groups in your experiments to provide a baseline for comparison, ensuring that the results are not influenced by external factors.

Select Appropriate Statistical Tests: Choose statistical methods that match the type of data and experimental design. Common methods include t-tests, ANOVA, regression analysis, and non-parametric tests, depending on the data distribution and study design. Understand Data Types: Differentiate between categorical and continuous data, as well as independent and dependent variables. This understanding guides you in selecting the right analysis technique. Check Assumptions: Ensure that the statistical methods used meet their underlying assumptions, such as normality, homogeneity of variance, and independence of observations

I am not sure how many times this has come up in seminars across schools globally. A great statistics course is key for a good PhD program. Not every school has it but thanks to the internet we all have infinite resources to learn them and ensure that we use the right ones for our data.

While I value statistics, few practical devices I've worked on are successful if the answer cannot be observed simply. Just as Feynman is attributed with "if you can't explain it, you don't understand it" if the results require complicated analyses, I strive to ensure that I can explain it in a clear manner. I focus attention in my practice on the outcome. I strive to take the results and view / report them in a manner in which the customer needs the information. I look to answer: "can the diagnosis/therapy be executed/realized to the level of comfort and certainty of the user". The statistics should be scrutinized, but the output for medical / biotech decision making should be simple enough for the customer to take action.

Employ rigorous statistical methods to analyze data. Proper statistical techniques help in interpreting conflicting results and identifying patterns that lead to accurate conclusions.

Peer review is essential to both the scientific and engineering methods. Creating peer review in for-profit institutions can be challenging. Internal team members should be sought out for reviews. External reviewers need sufficient information and context to provide meaningful review. External review can be extremely valuable to my practice, but the amount of background is typically several times the amount of material that I need reviewed. Each problem I face is typically so unique to a given situation that reviewers need significant effort to be brought up to speed to perform a valuable review.

Engage with Peers Early: Share preliminary findings with colleagues or within informal discussion groups before submitting to journals. Early feedback can help refine your approach and identify potential weaknesses in your methodology or analysis. Select Appropriate Reviewers: When submitting to journals, consider the expertise of potential reviewers. Select individuals who are knowledgeable about your specific area of research, as they can provide targeted feedback. Present at Conferences: Actively participate in conferences and seminars to share your research findings

There is no getting around the peer review in scientific research. I wouldn't say all journals have a rigorous process. However for students, publishing in pre-prints might also be useful. Many journals have editors who are looking for good pre- prints for their journals and will send you an invitation to submit if yours is a good fit.

Have your research reviewed by experts in the field. Peer review provides an additional layer of scrutiny to ensure that interpretations are logical and backed by evidence.

If you dig deep into your conflicting results, there is an original observation underneath, waiting for you to discover it. Many labs force students to ignore this data and change hypothesis, but the best results are often along the pathway less travelled.

Stay Updated with Literature: Regularly read peer-reviewed journals, articles, and reviews in bioengineering to stay informed about new research findings, methodologies, and technologies. Set up alerts for relevant topics to receive updates on the latest publications. Attend Conferences and Workshops: Participate in industry conferences, workshops, and symposiums to network with experts and gain insights into cutting-edge research and techniques. These events often feature presentations on recent advancements and provide opportunities for discussion.

Stay updated with the latest research and methodologies in bioengineering. Continuous learning helps in refining techniques and tools to interpret data accurately and avoid outdated methods.