Intro

I am a researcher working at the interface of substance abuse treatment, artificial intelligence, statistics and neuroscience.

Work

Multimodal predictive modeling of individual treatment outcome in cocaine dependence with combined neuroimaging and behavioral predictors (with D Martinez, KM Carpenter and EV Nunes)

Positron emission tomography (PET) has shown that the [11C]raclopride signal in the ventral striatum is significantly associated with treatment success in a positive reinforcement contingency management program. We hypothesize that this signal can be used to predict treatment outcome at an individual level. Methods Using logistic regression models, we show that individual treatment response can be predicted with a substantial degree of accuracy (cross-validated correct rate = 82%). Results Incorporating information from other regions-of-interest (ROIs) in the striatum does not improve predictive performance, except for a small improvement for adding the posterior caudate. The addition of baseline demographic variables does not improve predictive performance. Individual treatment responses can be also modeled with average clinic attendance as a behavioral predictor, and most of the predictive performance (83%) is reached by week 3 in the 24-week study. The combined model with both PET signal and clinic attendance record demonstrates a significant improvement of performance (93%), suggesting that neuroimaging and behavioral predictors comprise two distinct sources of information. Conclusions These results suggest that a biomarker for the treatment of substance use disorders may be built with only a few multimodal predictors, and pose hypotheses for the mechanisms of individual variations in behavioral outcome.

Predictive modeling and nonlinear treatment effects in a multicenter, randomized controlled trial of methylphenidate in smoke cessation intervention (with L Covey, M Hu, FR Levin, and EV Nunes)

There is evidence suggesting that cigarette smoking is more common in children and adults with attention-deficit/hyperactivity disorder (ADHD) compared to the general population. Whether and for whom treating the co-morbid ADHD would be part of an effective smoking cessation intervention remains uncertain. Leveraging the NIDA CTN28 trial data, we built and evaluated the performance of linear and nonlinear predictive models, and calculated the probability that a patient would stop smoking when treated either with methylphenidate or with a placebo. Results Consistent with previous findings, we recapitulated a number of covariates of treatment effects in the linear model. Simulations using the nonlinear models revealed a significant nonlinear treatment effect for baseline ADHD severity, with methylphenidate superior to placebo for patients with high baseline ADHD severity, but placebo superior to methylphenidate for those with low baseline ADHD severity. The threshold for treatment efficacy was calculated and varied as a function of other demographic characteristics. Simulations also reaffirmed that methylphenidate would improve ADHD symptoms for patients with both high and low baseline severity. Conclusions Predictive modeling provided an intuitive and clinically tractable measure of treatment success incorporating possible nonlinear treatment effects, and suggested that the success of smoke cessation treatment in co-morbid ADHD patients might benefit from customization of treatment. This method can be used to improve future trial design and assist in personalizing treatment in other substance use disorders.

Semantic mapping in adults with Autism Spectrum Disorder and typically developing adults. (with BS Peterson and AJ Gerber)

Individuals with Autism Spectrum Disorder are known to exhibit a variety of subtle pathologies in their language development, especially of language involved in social communications. We hypothesize that a systematic and quantitative characterization of these pathologies may reveal pathophysiological mechanisms in social language generation in these patients. We obtained the descriptions of traits of close social contacts such as parents, close friends and relatives of 26 individuals with high-functioning Autism Spectrum Disorder and 150 control subjects. These descriptions are analyzed using Latent Semantic Indexing (LSI). Machine learning classification and regression algorithms are used to visualize and characterize the differences between the patients and the controls. The rank-reduced descriptions from patients exhibit a deficit in distribution along a few specific axes in high dimensions in the semantic space, and this difference in distribution can be exploited in classifying patients and controls. This difference is maintained across different kinds of social relationships, and is not related to the relative valence in the described trait. Visualizing the semantic network reveals a reduction in semantic connectivity, and a less "small-world" network from the descriptions from patients as compared to those from controls. Social language production and its deficits may be measured using natural language processing methods. These findings may be helpful in defining the pertinent cortical circuit elements in the pathogenesis of Autism Spectrum Disorder.

About

CV

Contact

xsl2101@columbia.edu
Phone: 646-776-6144
Fax: 646-774-6111
Unit 66, New York State Psychiatric Institute
1051 Riverside Dr.
New York, NY 10033

Elements

Text

This is bold and this is strong. This is italic and this is emphasized. This is superscript text and this is subscript text. This is underlined and this is code: for (;;) { ... }. Finally, this is a link.


Heading Level 2

Heading Level 3

Heading Level 4

Heading Level 5
Heading Level 6

Blockquote

Fringilla nisl. Donec accumsan interdum nisi, quis tincidunt felis sagittis eget tempus euismod. Vestibulum ante ipsum primis in faucibus vestibulum. Blandit adipiscing eu felis iaculis volutpat ac adipiscing accumsan faucibus. Vestibulum ante ipsum primis in faucibus lorem ipsum dolor sit amet nullam adipiscing eu felis.

Preformatted

i = 0;

while (!deck.isInOrder()) {
    print 'Iteration ' + i;
    deck.shuffle();
    i++;
}

print 'It took ' + i + ' iterations to sort the deck.';

Lists

Unordered

  • Dolor pulvinar etiam.
  • Sagittis adipiscing.
  • Felis enim feugiat.

Alternate

  • Dolor pulvinar etiam.
  • Sagittis adipiscing.
  • Felis enim feugiat.

Ordered

  1. Dolor pulvinar etiam.
  2. Etiam vel felis viverra.
  3. Felis enim feugiat.
  4. Dolor pulvinar etiam.
  5. Etiam vel felis lorem.
  6. Felis enim et feugiat.

Icons

Actions

Table

Default

Name Description Price
Item One Ante turpis integer aliquet porttitor. 29.99
Item Two Vis ac commodo adipiscing arcu aliquet. 19.99
Item Three Morbi faucibus arcu accumsan lorem. 29.99
Item Four Vitae integer tempus condimentum. 19.99
Item Five Ante turpis integer aliquet porttitor. 29.99
100.00

Alternate

Name Description Price
Item One Ante turpis integer aliquet porttitor. 29.99
Item Two Vis ac commodo adipiscing arcu aliquet. 19.99
Item Three Morbi faucibus arcu accumsan lorem. 29.99
Item Four Vitae integer tempus condimentum. 19.99
Item Five Ante turpis integer aliquet porttitor. 29.99
100.00

Buttons

  • Disabled
  • Disabled

Form