ENAR 2024 Educational Program | TUTORIALS

Tutorials are roughly 2 hours in length and focus on a particular topic or software package. The sessions are more interactive than a standard lecture, often encouraging participants active engagement and hands-on participation.

 

Monday, March 11 | 8:30 am – 10:15 am
T1 | Teaching RNA-velocity for Single-cell Trajectory Analysis and Discussing its Future Research Directions

Instructor:
Kevin Lin, PhD, Department of Biostatistics, University of Washington

Course Description:

Over the past decade, single-cell analyses in developmental biology have garnered tremendous interest from the bioinformatics, mathematics, and statistics communities. RNA velocity is a concept at the forefront of this area and has led to novel scientific findings regarding how cells develop and commit to various fates. These findings are made possible by marrying statistical ideas with core bioinformatics and cell biology concepts. However, this combination of various disciplines can intimidate many new researchers from entering this topic.

In this tutorial, we will introduce participants to the area of RNA velocity in a way that distills its complexity into its key ingredients and discusses its branching directions of ongoing research. The tutorial begins with a gentle introduction to single-cell data and the underlying bioinformatics principles of RNA velocity. Then, participants learn the statistical backbone of RNA velocity, experience coding demos demonstrating this method, and see what biological insights it unlocks. Finally, the tutorial discusses how current work has extended RNA velocity in bioinformatics, computational, statistical, or laboratory-technology-driven directions.

The course teaches participants about RNA velocity, a topic combining ideas in bioinformatics, dynamical systems, and statistics. Participants interested in biological research will conclude the tutorial with a concrete roadmap to assess if RNA velocity applies to their own research and, if so, how to deploy it on their data. Participants interested in bioinformatics research will conclude the tutorial by seeing how recent work has successfully extended RNA velocity by leveraging other omics, or new laboratory technology and computational tools. Participants interested in statistics research will conclude the tutorial by seeing which aspects of RNA velocity could benefit from statistical inference tools that currently do not exist.

Statistical/Programming Knowledge Required:
Basic familiarity with R and Python; and basic understanding of ordinary differential equations, the EM algorithm, manifold learning, and deep learning are desired.

Instructor Biography:

---

Monday, March 11 | 10:30 am – 12:15 pm
T2 | Statistical Methods for Neighborhood and Community Focused Impacts in Health (In)Equity Research

Instructor:
Loni Philip Tabb, PhD, Dornsife School of Public Health, Drexel University

Course Description:

Understanding the impact of neighborhoods and communities on health and social inequities continue to present as a public health challenge for researchers, policy makers, and (bio)statisticians alike. There is a myriad of statistical approaches that aim to truly understand the intersection of health and place and time – with an emphasis on statistical implications. This tutorial will focus on highlighting spatial and spatio-temporal methods for multi-level settings, where both frequentist and Bayesian statistical frameworks will be discussed. Empirical evidence of these various methods will be presented as motivating examples – many of which will focus on health and social inequities across the US. The intended audience for this tutorial includes graduate students, post-docs, and researchers interested in understanding more the statistical implications of incorporating a neighborhood and community based focused and understanding their impacts on health outcomes of interest, in addition to traditional individual-level risk factors.

Statistical/Programming Knowledge Required:
Completion of Graduate level linear and generalized linear modeling courses.

Instructor Biography:

---

Monday, March 11 | 1:45 pm – 3:30 pm
T3 | An Introductory Tutorial on Structural Equation Modeling in R with Lavaan

Instructor:
Douglas D. Gunzler, PhD, Center for Health Care Research & Policy, Case Western Reserve University at MetroHealth Medical Center

Course Description:

Structural equation modeling (SEM) is a multivariate technique that allows relationships among variables to be examined. SEM is often used in practice to model and test hypothesized causal relationships among observed and latent (unobserved) variables, including in analysis across groups. It can be viewed as the merging of a conceptual model, path diagram, confirmatory factor analysis and path analysis. SEM emphasizes model evaluation.

While there are specialized commercial software for SEM, R statistical programming language is widely accessible and includes Lavaan. Participants will experience a mixture of lecture, discussion and hands-on participation. We will introduce basic concepts, theory and SEM vocabulary, give real-world examples, and conduct sample analyses using Lavaan. While no knowledge of SEM is required, a fundamental understanding of regression analysis and R is recommended for participants taking this tutorial.

Since SEM is an extremely broad set of methods, only foundational topics will be introduced in this tutorial. This tutorial is divided into three parts to introduce the basic vocabulary, concepts and usages of SEM to course participants, present the underlying statistical theory and then provide applications with R.

Statistical/Programming Knowledge Required:
While no knowledge of structural equation modeling is required, a fundamental understanding of regression analysis and R is recommended for participants taking this tutorial.

Instructor Biography:

---

Monday, March 11 | 3:45 pm – 5:30 pm
CANCELLED T4 | Improving Data-Preprocessing and Machine Learning Workflows with Scikit-Learn’s ColumnTransformer and Pipeline Classes

Instructor:
Gul Inan, Department of Mathematics, Istanbul Technical University

Course Description:

Python’s scikit-learn is one of the commonly used machine learning libraries which provide a comprehensive set of tools for building and deploying several supervised and unsupervised machine learning algorithms. Among its many features, scikit-learn offers the ColumnTransformer and Pipelines classes, which are powerful tools for data pre-processing and organization of multiple steps in machine learning workflows.

The ColumnTransformer class in scikit-learn provides a flexible mechanism for applying different pre-processing approaches to specific features of a dataset. This flexibility is particularly useful when dealing with datasets that contain a mixture of numerical and categorical features and feature engineering tasks such as scaling for numerical features and one-hot encoding for categorical features is required.

Complementing the ColumnTransformer, the Pipeline module in scikit-learn allows users to chain multiple pre-processing approaches such as imputation of missing values prior to scaling the numerical features or encoding the categorical features. The Pipeline module further allows users to concatenate these pre-processing steps and machine learning algorithms into a single object. By encapsulating the entire workflow within a pipeline, users are able to apply data pre-processing steps to both training and test data without data leakage and are able to simplify model training and deployment.

In this tutorial, we will first provide an overview of the scikit-learn’s ColumnTransformer and Pipeline classes. We will then discuss how the ColumnTransformer class enables flexible pre-processing of different type of features including scaling numerical features, encoding categorical features, and handling missing values. Lastly, we will explore how the Pipelines class allows for the efficient chaining of pre-processing steps with several machine learning algorithms such as regression, classification, principal component analysis, and K-means clustering.

Statistical/Programming Knowledge Required:
Familiarity with feature pre-processing approaches such as scaling, normalization, one-hot encoding, and imputation. Familiarity with lasso regression, ridge regression, their logistic regression counterparts, principal component analysis, and k-means clustering algorithms. Prior experience with Python’s scikit-learn API and a computational document such as Jupyter Notebook or JupyterLab is required.

Instructor Biography:

Dr. Gul Inan is currently an Assistant Professor in the Department of Mathematics at Istanbul Technical University, Turkey. She obtained her B.Sc., M.Sc., and Ph.D. all in Statistics from Department of Statistics, Middle East Technical University, Turkey. She did post-doctoral research at School of Statistics, University of Minnesota and Department of Biostatistics, University of North Carolina-Chapel Hill. She is interested in teaching “Statistical Learning with Python” at undergraduate and graduate level and developing statistical learning algorithms and/or computational tools to analyze high-dimensional data.

---

Tuesday, March 12 | 1:45 pm – 3:30 pm
T5 | Mentoring in Statistical Writing: Tools for Teaching and Energizing the Statistical Writing of your Mentees

Instructors:
Nicole Dalzell, PhD, Department of Statistical Sciences at Wake Forest University
Tanya Garcia, PhD, Department of Biostatistics and a Provost Distinguished Leader at UNC Chapel Hill

Course Description:

There are many different mentoring roles that involve teaching mentees to write or refine scientific writing skills, but individuals with training in statistics may not have training in teaching others to write. The goal of the tutorial is to provide guidance, practical tips, and teaching tools for mentoring in statistical writing. We will also discuss ways to incorporate mentoring and teaching about scientific writing in statistics into your work with mentees in a sustainable way.

In this tutorial, we will:

• Emphasize the importance of working to refine and energize mentee writing.
• Identify key skills for writing a research paper and provide activities and tools to help mentees develop and hone these skills.
• Discuss mentoring strategies with specific focus on improving mentee writing in statistics, keeping the individual needs and identities of each mentee in mind.

Statistical/Programming Knowledge Required:
None

Instructor Biographies:

---

Tuesday, March 12 | 3:45 pm – 5:30 pm
T6 | Introduction to MSM with Applications in Drug Development

Instructors:
David A. James, Novartis
Sophie Sun, Novartis
Meng Cao, Novartis
Alex Ocampo, Novartis Pharma AG

Course Description:

Multistate models (MSMs) are statistical models that are used to analyze complex and dynamic processes in which individuals or objects transition between multiple states over time. They can be viewed as an extension to traditional survival analysis where only one event is of interest (e.g. death or progression). The models allow for the estimation of transition probabilities between states and can be used to evaluate the effects of interventions or covariates on these probabilities. With appropriate set up with states and transitions, the models allow a more comprehensive and in-depth exploration of the dynamics and interdependence of patients’ and product’s characteristics on patients’ trajectories.

In this tutorial we introduce MSMs and describe a workflow using MSM for analyzing patients’ disease states collected from clinical studies. In the first part, we introduce basic MSM concepts such as what is a MSM and what are states, etc. In addition, we illustrate key points about the analyses and design decisions in the estimand framework via several examples from drug development, which highlights the flexibility of MSMs to be applied in a variety of settings. In the second part, we describe analytic examples for applying these techniques to data collected from clinical trials, and they include (1) descriptive techniques that extend Kaplan-Meier survival estimation to multiple state probabilities, (2) semi-parametric and parametric transition models for fitting patients’ trajectories along the state space, and (3), provide interpretation of the results from multistate model, as well as their relationship to the traditional time-to-event models (e.g. how to obtain restricted mean time, probabilities of being at each state on specific time point). Finally, time permitting, we will show R code fitting the MSMs in the case studies.

Statistical/Programming Knowledge Required:
Basic knowledge of survival analysis, including, e.g., KM, log-rank test, and Cox model.

Instructor Biographies: