What Does a Data Scientist Do? A Guide to the Role and Its Impact

A data scientist transforms raw information into decisions that can be tested, repeated, and improved. In research settings, the role combines statistical reasoning, programming, and domain knowledge to move from a question to a usable answer. This article explains the core activities of a data scientist, from data exploration to model deployment, with attention to how these activities apply in life sciences and related fields. It also provides a decision tree to help you determine whether a data science role fits your current skills and working preferences.

At a Glance

The table below summarizes the main phases of data science work, the typical outputs at each phase, and the records you should keep to verify that the work is sound.

Defining the Data Scientist Role

A data scientist is someone who uses quantitative methods to answer questions that matter to an organization or a field of study. The work is not limited to writing code or running algorithms. It includes deciding which data to trust, how to measure a concept, and what limitations to report alongside any finding.

In research contexts, the data scientist often works alongside domain experts who know the biology, chemistry, or clinical context. The data scientist contributes the technical skill to handle large or messy datasets, the statistical judgment to avoid false conclusions, and the communication ability to make results understandable. The Research Data Framework from the National Institute of Standards and Technology describes how research data should be managed across its lifecycle, which is a useful reference for anyone building a data workflow from scratch.

The role sits at the intersection of several disciplines. Biotechnology, bioinformatics, and bioinformation are linked fields where data science methods accelerate discovery in healthcare, agriculture, and food production. A review of this educational path in Bioinformation describes how biochemistry, genetics, and computer science combine to support drug development, vaccine design, and seed improvement. A data scientist in these settings must understand both the computational tools and the biological questions they serve.

Core Responsibilities Across the Data Lifecycle

Data Collection and Curation

The first responsibility of a data scientist is to obtain data that can actually answer the question at hand. This involves identifying sources, checking their reliability, and documenting any limitations. Public repositories such as NCBI Literature Resources and PubMed provide access to published research and associated data that can support new analyses.

Data curation is the process of organizing raw material into a usable form. This includes renaming variables consistently, converting units, handling missing values, and removing duplicates. The goal is a dataset that another person can open and understand without asking you for clarification. A data dictionary that defines each variable, its units, and its allowed values is a core deliverable at this stage.

Data Exploration and Hypothesis Generation

Once the data is clean, the data scientist explores it to understand its structure. This means calculating summary statistics, plotting distributions, and looking for patterns or anomalies. Exploration is not confirmatory analysis. It is the stage where you notice that a variable has an unexpected range, that a subgroup behaves differently, or that a relationship appears stronger than expected.

The Experimental Design Assistant from NC3Rs is a tool that helps researchers plan experiments before collecting data. Using such planning tools reduces the risk that exploration will lead to false hypotheses. A data scientist should be able to distinguish between patterns that emerge from careful exploration and patterns that are artifacts of poor data collection.

Statistical Modeling and Machine Learning

Modeling is the phase where the data scientist builds a formal representation of the relationship between variables. Statistical models such as regression estimate the strength of associations and provide confidence intervals. Machine learning models such as random forests or neural networks focus on prediction accuracy, often at the cost of interpretability.

The choice between statistical modeling and machine learning depends on the question. If the goal is to understand which factors drive an outcome, an interpretable statistical model is usually better. If the goal is to classify new cases accurately, a machine learning model may perform better even if its internal logic is harder to explain.

A systematic review of data science methodologies in supply chain and logistics research shows how the same modeling techniques are applied across efficiency, resilience, and sustainability questions. The review classifies the models used and identifies gaps in current practice. This illustrates that data science methods are transferable across domains, but the specific choices depend on the context.

Model Validation and Reproducibility

A model that performs well on the data used to build it may fail on new data. Validation is the process of testing the model on data it has not seen during training. Common approaches include splitting the data into training and test sets, using cross-validation, or collecting a separate validation dataset.

Reproducibility is a related but distinct concern. A reproducible workflow is one that another researcher can run to obtain the same results. This requires version control for code, clear documentation of software versions, and a record of the exact data used. The Ten simple rules for building and maintaining a responsible data science workflow provides practical guidance on structuring work so that it can be audited and repeated.

Deployment and Communication

Deployment is the process of moving a model from a research notebook into a tool that others can use. This might be a dashboard, a report, or an automated pipeline that processes new data as it arrives. Deployment requires attention to software engineering practices such as testing, logging, and error handling.

Communication is the final responsibility. A data scientist must explain what the model does, what it does not do, and what assumptions underlie it. This is especially important when the results will inform decisions about health, safety, or resource allocation. The S-RACE platform described in Recenti progressi in medicina is an example of a cloud-based system designed to move AI research into clinical practice while addressing data quality and governance. The platform includes tools for both clinicians and data scientists, recognizing that both groups need to understand the system for it to work.

The Data Science Workflow in Practice

Step 1: Define the Question

Every data science project starts with a question that is specific enough to guide the work. A vague question such as "what affects crop yield" is not actionable. A better question is "does soil nitrogen level at planting predict final yield in wheat under rainfed conditions." The question determines which data to collect, which methods to use, and how to interpret the results.

Step 2: Acquire and Clean the Data

Data acquisition involves locating sources, downloading or extracting the relevant records, and checking that the data matches the question. Cleaning is the process of correcting errors, filling or removing missing values, and standardizing formats. This step often takes more time than any other part of the workflow.

Step 3: Explore and Visualize

Exploration produces the first insights. Plotting the data reveals distributions, outliers, and relationships that summary statistics miss. This is also the stage where you check whether the data meets the assumptions of the methods you plan to use. For example, a linear model assumes a roughly linear relationship between predictors and outcome, which a scatterplot can confirm or contradict.

Step 4: Build and Validate the Model

Model building is iterative. You start with a simple model, check its performance, and add complexity only when it improves results on validation data. The Path4Drug workflow described in Frontiers in Pharmacology shows how a data pipeline can connect drug targets to biological pathways using public repositories. The workflow was applied to withdrawn drugs and drugs with safety warnings to identify pathways linked to toxicity. This example demonstrates how a well-structured pipeline can be reused across different compounds and questions.

Step 5: Document and Report

Documentation is not an afterthought. It is the record that allows others to verify your work and build on it. A complete report includes the question, the data sources, the cleaning steps, the modeling choices, the validation results, and the limitations. The EQUATOR Network provides reporting guidelines for health research that help authors include the information readers need to assess study quality.

Tools and Technologies Used by Data Scientists

Data scientists rely on a range of tools that support different parts of the workflow. Programming languages such as Python and R are the most common for analysis. Both have extensive libraries for statistics, machine learning, and visualization.

Workflow managers help organize complex analyses into reproducible pipelines. The TE_Bench workflow described in Mobile DNA uses Snakemake to automate the benchmarking of transposable element annotation pipelines. The workflow can simulate DNA sequences with known transposable elements or accept real data, allowing researchers to compare how well different tools detect these elements. This example shows how workflow managers support systematic comparison and reuse.

Cloud platforms provide computing resources and collaboration features that are difficult to replicate on a single machine. The mdx platform described in the 2022 IEEE conference proceedings is a nationwide cloud service in Japan that supports researchers in materials science, life science, and other fields. The platform provides secure access to supercomputers and edge devices, enabling analyses that would be impractical on local hardware.

Specialized tools exist for particular domains. QIIME 2, described in Nature Biotechnology, is an open-source platform for microbiome data science. It supports reproducible analysis of microbial community data and is extensible through plugins. Researchers studying the microbiome can use QIIME 2 to move from raw sequencing data to publication-ready figures while maintaining a complete record of the analysis steps.

Data Science in Life Sciences and Research

Genomics and Precision Medicine

Data science is central to genomics, where datasets contain millions of measurements per sample. The Singapore National Precision Medicine Strategy described in Nature Genetics is a 10-year initiative to integrate genomic, lifestyle, health, social, and environmental data for up to one million individuals. The strategy recognizes that routine adoption of precision medicine requires addressing social, ethical, legal, and regulatory barriers alongside technical ones. Data scientists working in this area must handle large-scale genomic-phenotypic databases while ensuring that the results are clinically meaningful.

Long-read sequencing generates data that requires specialized analysis workflows. A trio-based long-read sequencing workflow described in the Journal of Human Genetics was used to identify a pathogenic transposable element insertion in a patient with Simpson-Golabi-Behmel syndrome. The workflow prioritized candidate structural variants by narrowing tens of thousands of loci to fewer than 20 for manual inspection. This case shows how data science methods can directly contribute to diagnosis of rare diseases.

Digital Pathology and Medical Imaging

Pathology is becoming increasingly digital, and data scientists are needed to develop and validate algorithms that assist pathologists. A systematic review of foundation models and AI agents in digital pathology found that these models perform well on diagnostic tasks after fine-tuning, with area under the curve values ranging from 0.85 to 0.97. However, only 14.3 percent of the included studies conducted multicenter validation, and fewer than 5 percent were prospective. The review identifies data heterogeneity, limited explainability, insufficient workflow-level validation, and regulatory gaps as barriers to clinical adoption.

Surgical workflow recognition is another area where data science supports clinical practice. The LungRes80 dataset described in Medical Image Analysis contains 269,806 video frames from 80 video-assisted thoracoscopic surgery cases. The associated model uses causal reasoning to recognize surgical phases and quantify the cost of intraoperative distraction. This work aims to support monitoring and early warning during high-risk procedures.

Plant Phenotyping and Agriculture

Data science methods are being applied to plant phenotyping to measure traits such as height, canopy width, and leaf dimensions. The F2DMAS workflow described in Frontiers in Plant Science uses smartphone videos to reconstruct three-dimensional plant structure in complex backgrounds. The workflow achieved strong agreement with manual measurements, with R-squared values ranging from 0.90 to 0.99. This approach makes phenotyping more accessible because it uses consumer-grade equipment instead of specialized sensors.

Catalysis and Materials Research

Data science is also transforming chemistry and materials science. Embracing data science in catalysis research, published in Nature Catalysis, discusses how machine learning and data-driven methods are being integrated into catalyst discovery and optimization. Researchers in this field use data science to screen candidate materials, predict performance, and identify promising directions for experimental work.

Skills and Qualifications for Data Science Roles

Technical Skills

A data scientist needs proficiency in at least one programming language, typically Python or R. Statistical knowledge is essential, including hypothesis testing, regression, and experimental design. Machine learning skills are increasingly expected, including the ability to train, validate, and interpret models.

Data manipulation skills are equally important. This includes working with structured data in tables, unstructured data such as text or images, and large datasets that require distributed computing. Version control using tools such as Git is a standard expectation in professional settings.

Domain Knowledge

Domain knowledge is what separates a data scientist who can produce technically correct results from one who can produce useful results. In life sciences, this means understanding biological concepts, experimental design, and the limitations of measurement instruments. A data scientist who understands the domain can ask better questions, interpret results in context, and communicate findings to domain experts.

The bibliometric analysis of data science research presented at ICSTCEE 2023 shows how the field has grown over the past decade, with increasing publication output and expanding research networks. The analysis identifies major research hubs and collaboration patterns, providing a map of where data science expertise is concentrated.

Communication and Collaboration

Data scientists must communicate with people who do not share their technical background. This requires the ability to explain statistical concepts in plain language, to present results visually, and to translate findings into recommendations. Collaboration skills are equally important because data science projects typically involve multiple stakeholders with different expertise.

The study on applying data science to improve student learning outcomes demonstrates how data science methods can support decision-making in education. The study used machine learning to identify factors such as attendance and engagement that correlate with learning outcomes, allowing educators to support at-risk students. This example shows how data scientists must work with domain experts to ensure that models address real needs.

Common Failure Patterns in Data Science Work

Overfitting the Training Data

Overfitting occurs when a model learns the noise in the training data instead of the underlying pattern. The model performs well on the data used to build it but poorly on new data. This is one of the most common errors in machine learning. Prevention requires validation on held-out data and a preference for simpler models when performance is similar.

Ignoring Data Provenance

Data provenance is the record of where data came from and how it was processed. Without provenance, it is impossible to verify results or troubleshoot errors. A common failure is to receive a dataset from a collaborator without documenting its original source, cleaning steps, or known limitations. This creates risk that downstream analyses will be based on incorrect assumptions.

Confusing Correlation with Causation

Data science methods can identify associations, but they cannot establish causation without appropriate study design. A common failure is to interpret a correlation as evidence that one variable causes another. The debate over food addiction described in Frontiers in Psychiatry illustrates how difficult it can be to establish causal relationships in nutrition research. The scientific community remains divided over whether certain foods are addictive, with arguments about correlation versus causation playing a central role in the controversy.

Poor Documentation

A workflow that is not documented is not reproducible. Even if the results are correct, the lack of documentation prevents others from verifying them or building on them. Poor documentation also makes it difficult to revisit a project months later because the analyst may not remember what was done or why.

Ignoring Model Limitations

Every model has limitations, and a responsible data scientist reports them. This includes the range of data on which the model was validated, the assumptions it makes, and the situations where it may fail. Ignoring limitations can lead to overconfident decisions based on results that do not generalize.

Ethical and Governance Considerations

Algorithmic Accountability

Data science methods are increasingly used in decisions that affect individuals and communities. The discussion of algorithmic accountability in Philosophical Transactions of the Royal Society A outlines concerns about how algorithms might be held to account in areas affecting the public sphere. The paper argues for greater transparency, monitoring of outcomes, and improved governance. It also recommends diversity in the workforce, ethics training, and codes of conduct for data scientists.

Responsible Data Science Workflows

A responsible data science workflow incorporates ethical considerations at every stage, not as an afterthought. This includes asking whether the data was collected with appropriate consent, whether the analysis could produce biased results, and whether the deployment could cause harm. The Ten simple rules for building and maintaining a responsible data science workflow provides practical guidance for structuring work in a way that supports accountability.

Public Trust and Scientific Integrity

Public trust in science depends on the integrity of researchers and the transparency of their methods. A perspective in Molecular Cell suggests that scientists' reluctance to engage the public on controversial ethical issues may contribute to moderate public distrust of scientists' motivations. The authors propose a Scientist's Oath to communicate the ideals that guide scientific work. Data scientists, like other researchers, have a responsibility to be transparent about their methods and honest about limitations.

Ethical Obligations in Research

Scientists have ethical obligations that extend beyond the technical quality of their work. The discussion of scientists' ethical obligations and social responsibility for nanotechnology research in Science and Engineering Ethics addresses how researchers should consider the broader implications of their work. Data scientists face similar questions about how their models might be used and misused.

Decision Tree for Assessing Fit with a Data Science Role

Use the following questions to determine whether a data science role fits your skills and preferences.

Question 1: Do you enjoy working with data?

Data science involves spending substantial time cleaning, exploring, and analyzing data. If you find this work tedious, the role will be a poor fit regardless of your technical skills. If you enjoy discovering patterns and solving puzzles in data, the role may suit you.

Question 2: Are you comfortable with programming?

Programming is a core skill for data scientists. You do not need to be a software engineer, but you need to be comfortable writing code to manipulate data, build models, and automate tasks. If you have no interest in programming, consider whether a related role such as a statistician or domain expert might be a better fit.

Question 3: Do you have or can you develop statistical knowledge?

Data science requires a solid foundation in statistics. This includes understanding probability, hypothesis testing, regression, and experimental design. If you are willing to study these topics, you can build the necessary knowledge. If you have no interest in statistical reasoning, the role will be difficult.

Question 4: Can you communicate technical results to nontechnical audiences?

Data scientists must explain their work to people who do not share their technical background. This includes presenting results, writing reports, and answering questions. If you prefer to work alone and avoid communication, consider whether you can develop these skills or whether a different role would be more suitable.

Question 5: Are you comfortable with uncertainty?

Data science rarely produces definitive answers. Models have limitations, data has gaps, and conclusions can change as new information becomes available. If you need certainty in your work, the ambiguity of data science may be frustrating.

Question 6: Do you care about the domain you would work in?

Data science methods are transferable, but the work is more satisfying and more effective when you care about the questions you are answering. If you are interested in life sciences, agriculture, or healthcare, a data science role in these fields can be rewarding. If you have no interest in the domain, consider whether the technical work alone is enough to sustain your engagement.

Records and Measurements for Data Science Work

What to Record

A data science project should maintain records of the data sources, cleaning steps, modeling choices, and validation results. The data dictionary is a core record that defines each variable and its properties. The analysis notebook or script provides a chronological record of the work. The model card summarizes the model's purpose, performance, and limitations.

How to Measure Success

Success in data science is measured by the usefulness of the results, not by the complexity of the methods. A simple model that answers the question and can be explained is more valuable than a complex model that performs slightly better but cannot be understood. Validation metrics such as accuracy, precision, recall, and area under the curve provide quantitative measures of model performance, but they must be interpreted in the context of the question.

When to Escalate

Data scientists should escalate concerns when they encounter data quality issues that cannot be resolved, when model performance is inadequate for the intended use, or when ethical concerns arise about how the results might be used. Escalation criteria should be defined at the start of a project so that problems are addressed promptly instead of discovered late.

Limitations of Data Science Methods

Data Quality Limits

Data science methods cannot compensate for poor data quality. If the data is biased, incomplete, or measured with error, the results will reflect those problems. The Research Data Framework from NIST emphasizes the importance of managing data quality across the research lifecycle.

Generalization Limits

Models are validated on specific datasets, and their performance may not generalize to other populations or settings. A model trained on data from one region may not work in another region with different environmental conditions. A model trained on data from one patient population may not work for another population with different genetic backgrounds.

Interpretability Limits

Some machine learning models are difficult to interpret, making it hard to understand why they make particular predictions. This is a concern in regulated settings where decisions must be explained. The review of foundation models in digital pathology identifies limited explainability as one of the barriers to clinical adoption.

Causal Inference Limits

Observational data can reveal associations, but establishing causation requires careful study design. The discussion of brain imaging in Cambridge Quarterly of Healthcare Ethics highlights the debate over whether brain scans can reliably establish causal connections between brain function and behavior. Similar debates occur in many fields where data science methods are applied to observational data.

Professional Escalation Criteria

Data scientists should escalate concerns to supervisors, collaborators, or relevant authorities when they encounter situations that exceed their expertise or authority. Specific escalation criteria include:

• Data quality problems that cannot be resolved with available resources
• Model performance that is inadequate for the intended use
• Evidence that the analysis could produce harmful outcomes
• Requests to misrepresent results or hide limitations
• Uncertainty about regulatory requirements or ethical obligations

The algorithmic accountability discussion in Philosophical Transactions of the Royal Society A argues that all regulators need to address the challenges posed by changing technology. Data scientists have a role in raising concerns when they see potential harms, even if the concerns are uncomfortable.

Frequently Asked Questions

What is the difference between a data scientist and a statistician?

A statistician focuses on statistical theory and methods, with emphasis on experimental design, hypothesis testing, and inference. A data scientist typically has a broader skill set that includes programming, machine learning, and data engineering. In practice, the roles overlap substantially, and the distinction varies by organization. A statistician might focus on designing a study and analyzing the results, while a data scientist might build a pipeline that processes data continuously and deploys a model for ongoing use.

Do I need a PhD to become a data scientist?

A PhD is not required for all data science roles, but it is common in research settings. A PhD demonstrates the ability to conduct independent research, which is valuable for roles that involve developing new methods or working on complex problems. For roles that focus on applying existing methods, a master's degree or equivalent experience may be sufficient. The bibliographic record on typical responsibilities and qualifications for data scientists in the Journal of Library Science in China discusses the education and qualifications expected for the role.

What programming languages should I learn for data science?

Python and R are the most widely used languages in data science. Python has a large ecosystem of libraries for machine learning, data manipulation, and visualization. R is strong in statistical analysis and has excellent tools for reproducible research. Learning both is valuable, but starting with one and building depth is more practical than trying to learn both simultaneously.

How much mathematics do I need for data science?

A solid foundation in linear algebra, calculus, and probability is important for understanding how machine learning methods work. You do not need to derive every formula from first principles, but you need enough mathematical understanding to make informed choices about methods and to interpret results correctly. The level of mathematics required depends on the role. Research roles that develop new methods require more mathematics than application roles that use existing tools.

What is the difference between machine learning and data science?

Machine learning is a subfield of data science that focuses on algorithms that learn patterns from data. Data science is broader and includes data collection, cleaning, exploration, statistical analysis, and communication. A data scientist may use machine learning methods, but the role also involves many activities that are not machine learning, such as designing experiments, cleaning data, and presenting results.

How do I know if a data science role is right for me?

Use the decision tree in this article to assess your fit. Consider whether you enjoy working with data, whether you are willing to learn programming and statistics, whether you can communicate technical results, and whether you are comfortable with uncertainty. Also consider whether you care about the domain you would work in. If you answer yes to most of these questions, a data science role may be a good fit.

What are the most common mistakes made by new data scientists?

The most common mistakes include overfitting models to training data, ignoring data provenance, confusing correlation with causation, and failing to document work. New data scientists often spend too much time on complex models before mastering basic data cleaning and exploration. They may also present results without stating limitations, which can mislead decision-makers.

How do data scientists ensure their work is reproducible?

Reproducibility requires version control for code, documentation of software versions, and a clear record of data sources and cleaning steps. Workflow managers such as Snakemake or Nextflow can automate pipelines so that the same steps are applied consistently. The TE_Bench workflow and QIIME 2 platform are examples of tools designed to support reproducible analysis.

Related Articles

• Research Scientist Careers in Biotechnology
• Research Statement for Bioinformatics Applications
• Biomedical Science Research Building
• Genome Research Impact Factor
• Scientific Data Provenance For Reproducible Research

References and Further Reading

• Research Data Framework. National Institute of Standards and Technology.
• EQUATOR Network. EQUATOR Network.
• Experimental Design Assistant. NC3Rs.
• NCBI Literature Resources. National Center for Biotechnology Information.
• PubMed. National Library of Medicine.
• A Scientist's Oath.. Molecular cell, 2018.
• La piattaforma S-RACE: una soluzione cloud per dati sanitari real-world, che guida la traslazione clinica e la governance responsabile della intelligenza artificiale.. Recenti progressi in medicina, 2025.
• Algorithmic accountability.. Philosophical transactions. Series A, Mathematical, physical, and engineering sciences, 2018.
• Heroes of the Engram.. The Journal of neuroscience : the official journal of the Society for Neuroscience, 2017.
• The battle over "food addiction".. Frontiers in psychiatry, 2025.
• The Singapore National Precision Medicine Strategy.. Nature genetics, 2023.
• Biotechnology, Bioinformatics and Bioinformation in an Autobiography.. Bioinformation, 2020.
• Naturalizing Responsibility.. Cambridge quarterly of healthcare ethics : CQ : the international journal of healthcare ethics committees, 2016.
• Ten simple rules for building and maintaining a responsible data science workflow.. 2024.
• Path4Drug: Data Science Workflow for Identification of Tissue-Specific Biological Pathways Modulated by Toxic Drugs.. 2021.
• TE_Bench: a foundational benchmarking workflow for transposable element annotation pipelines.. 2026.
• F2DMAS: a smartphone video-based 3D phenotyping workflow for potted plants in complex backgrounds.. 2026.
• A trio-based long-read sequencing workflow identifies a pathogenic transposable element insertion in a previously undiagnosed patient.. 2026.
• LungRes80: Towards tangled surgical workflow recognition in video-assisted thoracoscopic surgery.. 2026.
• Foundation Models and AI Agents in Digital Pathology Imaging: A Systematic Review of Integration into the Clinical Workflow and Implementation Challenges.. 2026.
• Embracing data science in catalysis research. Nature Catalysis, 2024.
• Reproducible, interactive, scalable and extensible microbiome data science using QIIME 2. Nature Biotechnology, 2019.
• Data science and big data analytics: a systematic review of methodologies used in the supply chain and logistics research. Annals of Operations Research, 2023.
• Bibliometric Analysis of Data Science Research: A Decade of Insights from Web of Science. 2023 Fourth International Conference on Smart Technologies in Computing, Electrical and Electronics (ICSTCEE), 2023.
• mdx: A Cloud Platform for Supporting Data Science and Cross-Disciplinary Research Collaborations. 2022 IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech), 2022.
• Applying Data Science to Analyze and Improve Student Learning Outcomes in Educational Environments. Information Technologies in Environmental Engineering, 2024.
• Mapping the Evolution of Social Research and Data Science on 30 Years of Social Indicators Research. Social Indicators Research, 2020.
• Typical Responsibilities, Key Qualifications and Higher Education for Data Scientist. Journal of Library Science in China, 2021.
• Scientists’ Ethical Obligations and Social Responsibility for Nanotechnology Research. Science and Engineering Ethics, 2016.

This article is educational and does not replace institutional policy, professional advice, or applicable safety and regulatory requirements.